Surrogate Modeling Techniques Research Articles

Uncertainty quantification and sensitivity analysis of neuron models with ion concentration dynamics.

This paper provides a comprehensive and computationally efficient case study for uncertainty quantification (UQ) and global sensitivity analysis (GSA) in a neuron model incorporating ion concentration dynamics. We address how challenges with UQ and GSA in this context can be approached and solved, including challenges related to computational cost, parameters affecting the system's resting state, and the presence of both fast and slow dynamics. Specifically, we analyze the electrodiffusive neuron-extracellular-glia (edNEG) model, which captures electrical potentials, ion concentrations (Na+, K+, Ca2+, and Cl-), and volume changes across six compartments. Our methodology includes a UQ procedure assessing the model's reliability and susceptibility to input uncertainty and a variance-based GSA identifying the most influential input parameters. To mitigate computational costs, we employ surrogate modeling techniques, optimized using efficient numerical integration methods. We propose a strategy for isolating parameters affecting the resting state and analyze the edNEG model dynamics under both physiological and pathological conditions. The influence of uncertain parameters on model outputs, particularly during spiking dynamics, is systematically explored. Rapid dynamics of membrane potentials necessitate a focus on informative spiking features, while slower variations in ion concentrations allow a meaningful study at each time point. Our study offers valuable guidelines for future UQ and GSA investigations on neuron models with ion concentration dynamics, contributing to the broader application of such models in computational neuroscience.

Impact of urban form and street infrastructure on pedestrian-motorist collisions.

This study examines the impact of urban form and street infrastructure on pedestrian safety in Atlanta, Georgia, and Boston, Massachusetts. With a significant rise in pedestrian fatalities over the past decade, understanding how cities' built environments influence safety is critical. We conducted geospatial analyses and statistical tests, revealing unique patterns in each city. Atlanta's sprawling, motorist-oriented layout is associated with increased pedestrian accidents, particularly at crosswalks, due to limited land use diversity, arterial roads, and streets with high speed limits and multiple lanes. In contrast, Boston's compact, pedestrian-oriented design leads to improved safety, featuring safer pedestrian crossings, greater land use diversity, reduced arterial roads and lower speed limits on single-lane streets. This study also highlights the importance of diverse urban forms and pedestrian infrastructure in shaping pedestrian safety. While population density and land use diversity impact accident rates, the presence of crosswalks and street configurations play crucial roles. Our findings underscore the urgency for urban planners to prioritize pedestrian safety through targeted interventions, such as enhancing crosswalks, reducing speed limits and promoting mixed land use. Future research should explore additional variables, alternative modelling techniques and non-linear approaches to gain a more comprehensive understanding of these complex relationships.

White men can't jump, but do they still get picked first? Race and player selection in the NBA draft, 1980-2021.

Despite excelling at recruiting Black players, studies have repeatedly produced evidence of racial discrimination in the National Basketball Association (NBA). Through this study, we re-examine the topic of racial discrimination within the NBA through an analysis of the Association's annual entry draft. Using a novel dataset, we statistically model the relationship between player race and draft pick number using pooled data from 1980 to 2021. Overall, we find only limited evidence of racial discrimination. These findings are generally robust to sub-sample analyses, alternative specifications of our race variable, and alternate statistical modeling techniques. However, analyses performed on sub-samples of draft picks that participated in the NBA combine-and for whom we have measurements of player athleticism-produce some evidence of racial discrimination. Through such models we estimate that Black players are picked roughly three picks later in the draft. We consider the implications of these findings for contemporary theorizing about racial discrimination in the NBA and more mainstream labor markets.

Energy absorption of a novel auxetic structure reinforced by embedding tubes

To enhance structural crashworthiness while achieving lightweight design, this paper proposes a novel configuration that embeds a thin-walled square tube into auxetic structure which can take full advantage of the interaction between the auxetic structure and embedding tubes. The finite element model of auxetic structures reinforced by embedding tubes is established and validated by the compression experiments. The crashworthiness performance of auxetic structure (AUX), auxetic structure reinforced by embedding tubes (AET), and auxetic structure-filled tubes (AFT) are compared to find that the AET have the highest energy absorption and specific energy absorption (SEA). Through parametric analysis, designing the angles between short and long beams (θ1, θ2) as 70° and 40° can contribute to the crashworthiness performance, respectively. A suitable range for long beam thickness lies between 1.4 mm and 1.6 mm, while a short beam thickness of around 1.2 mm is more effective. The specific configuration can enhance the SEA and reduces peak crushing force (PCF). Increasing the thickness of the embedded tube can improve the crashworthiness by sacrificing the weight. It is a good balance between the crashworthiness performance and weight by designing the tube thickness as 1.2 mm. An analytical model is also established to predict the energy absorption of auxetic structure reinforced by embedding tubes under axial impact loadings. The surrogate modeling technique and the NSGA-II algorithm are also employed to optimize the AET configuration. The results show that the SEA of the optimized structure can increase from 14.5 kJ/kg to 18.6 kJ/kg, while the PCF can be reduced from 164.6 kN to 124.3 kN. Therefore, the auxetic structure reinforced by embedding tubes can play a critical role in the engineering field to absorb energy.

Surrogate modeling with non-stationary-noise based Gaussian process regression and K-Fold ANN for systems featuring uneven sensitivity distribution

In the realm of aircraft design, there is a prevalent need for surrogate modeling techniques capable of efficiently and accurately modeling objects with high evaluation costs and uneven sensitivity distributions, such as those encountered in the nonlinear buckling of thin-walled structures and aerodynamics at transonic speeds. However, the non-stationary Gaussian Process Regression (GPR) model requires extensive hyperparameters or judicious assumptions, limiting its applications in such tasks, and the commonly used K-Fold methodology suffers from inherent instability. Addressing these issues, this paper proposes an active learning method for surrogate modeling in systems featuring uneven sensitivity distribution, integrating non-stationary-noise GPR and K-Fold Artificial Neural Network (ANN) methodology, namely NSN-GPR-KF. Instead of relying on a non-stationary kernel function, the algorithm regards noise as a non-stationary factor within a stationary GPR framework, with noise levels estimated via the K-Fold ANN methodology, allowing GPR to efficiently adapt to non-stationary systems while avoiding the aforementioned requirements for non-stationary GPR. Case validations, performed on two test functions with pronounced uneven spatial sensitivity, demonstrate that NSN-GPR-KF outperforms commonly used stationary GPR and pure K-Fold methodology. It synthesizes the exploration capabilities of GPR with the exploitation proficiency of the K-Fold methodology, achieving comparable global accuracy with reduced sample sizes—up to 26% and 22% less than the other two algorithms, respectively. The algorithm was successfully applied to predict the buckling load of thin-walled pipe beams, demonstrating accelerated convergence and approximately 25% greater accuracy compared to pure K-Fold model. These results suggest that NSN-GPR-KF can serve as an efficient and precise modeling tool for engineering tasks with high evaluation costs and complex sensitivity distributions.

Insights from a Comprehensive Capacity Expansion Planning Modeling on the Operation and Value of Hydropower Plants under High Renewable Penetrations

This paper presents a quantitative assessment of the value of hydroelectric power plants (HPPs) in power systems with a significant penetration of variable renewable energy sources (VRESs). Through a capacity expansion planning (CEP) model that incorporates a detailed representation of HPP operating principles, the study investigates the construction and application of HPP rule curves essential for seasonal operation. A comparative analysis is also conducted between the proposed rule curve formulation and alternative modeling techniques from the literature. The CEP model optimizes installed capacities per technology to achieve predefined VRES penetration targets, considering hourly granularity and separate rule curves for each HPP. A case study involving twelve reservoir hydropower stations and two open-loop pumped hydro stations is examined, accounting for standalone plants and cascaded hydro systems across six river basins. The study evaluates the additional generation and storage required to replace the hydropower fleet under high VRES penetration levels, assessing the resulting increases in total system cost emanating from introducing such new investments. Furthermore, the study approximates the storage capabilities of HPPs and investigates the impact of simplified HPP modeling on system operation and investment decisions. Overall, the findings underscore the importance of reevaluating hydro rule curves for future high VRES penetration conditions and highlight the significance of HPPs in the energy transition towards carbon neutrality.

Stochastic emulation with enhanced partial‐ and no‐replication strategies for seismic response distribution estimation

AbstractModern performance earthquake engineering practices frequently require a large number of time‐consuming non‐linear time‐history simulations to appropriately address excitation and structural uncertainties when estimating engineering demand parameter (EDP) distributions. Surrogate modeling techniques have emerged as an attractive tool for alleviating such high computational burden in similar engineering problems. A key challenge for the application of surrogate models in earthquake engineering context relates to the aleatoric variability associated with the seismic hazard. This variability is typically expressed as high‐dimensional or non‐parametric uncertainty, and so cannot be easily incorporated within standard surrogate modeling frameworks. Rather, a surrogate modeling approach that can directly approximate the full distribution of the response output is warranted for this application. This approach needs to additionally address the fact that the response variability may change as input parameter changes, yielding a heteroscedastic behavior. Stochastic emulation techniques have emerged as a viable solution to accurately capture aleatoric uncertainties in similar contexts, and recent work by the second author has established a framework to accommodate this for earthquake engineering applications, using Gaussian Process (GP) regression to predict the EDP response distribution. The established formulation requires for a portion of the training samples the replication of simulations for different descriptions of the aleatoric uncertainty. In particular, the replicated samples are used to build a secondary GP model to predict the heteroscedastic characteristics, and these predictions are then used to formulate the primary GP that produces the full EDP distribution. This practice, however, has two downsides: it always requires minimum replications when training the secondary GP, and the information from the non‐replicated samples is utilized only for the primary GP. This research adopts an alternative stochastic GP formulation that can address both limitations. To this end, the secondary GP is trained by measuring the square of sample deviations from the mean instead of the crude sample variances. To establish the primitive mean estimates, another auxiliary GP is introduced. This way, information from all replicated and non‐replicated samples is fully leveraged for estimating both the EDP distribution and the underlying heteroscedastic behavior, while formulation accommodates an implementation using no replications. The case study examples using three different stochastic ground motion models demonstrate that the proposed approach can address both aforementioned challenges.

Proof of concept of physiologically based pharmacokinetic modelling in paediatric acute lymphoblastic leukaemia.

Physiologically based pharmacokinetic (PBPK) modelling is an alternative modelling technique that is increasingly used in pharmacokinetics. Due to its nature, it can be complementarily employed to population pharmacokinetics, especially when it comes to small population size. Here, we report the proof of concept of its application to accurately describe the pharmacokinetics of a recombinant L-asparaginase in paediatric patients with acute lymphoblastic leukaemia. Data from two randomized, double-blind, phase II/III clinical studies (MC-ASP.4/ALL; MC-ASP.5/ALL) were included to setup and evaluate the final model, respectively. Final population values for basic pharmacokinetic parameters were calculated (clearance: 0.0569 L/h/19.5 kg, volume of distribution: 1.251 L, half-life: 18.5 h, trough concentration: 140.9 IU/L). Pharmacokinetic parameter prediction as well as predictive performance of the model proofed to be comparable to a separately developed population pharmacokinetic model with 13% deviation in predicted median L-asparaginase trough levels. To the best of our knowledge, this is the first whole-body PBPK model of a non-antibody therapeutic protein.

Pseudo-value regression trees.

This paper presents a semi-parametric modeling technique for estimating the survival function from a set of right-censored time-to-event data. Our method, named pseudo-value regression trees (PRT), is based on the pseudo-value regression framework, modeling individual-specific survival probabilities by computing pseudo-values and relating them to a set of covariates. The standard approach to pseudo-value regression is to fit a main-effects model using generalized estimating equations (GEE). PRT extend this approach by building a multivariate regression tree with pseudo-value outcome and by successively fitting a set of regularized additive models to the data in the nodes of the tree. Due to the combination of tree learning and additive modeling, PRT are able to perform variable selection and to identify relevant interactions between the covariates, thereby addressing several limitations of the standard GEE approach. In addition, PRT include time-dependent effects in the node-wise models. Interpretability of the PRT fits is ensured by controlling the tree depth. Based on the results of two simulation studies, we investigate the properties of the PRT method and compare it to several alternative modeling techniques. Furthermore, we illustrate PRT by analyzing survival in 3,652 patients enrolled for a randomized study on primary invasive breast cancer.

Investigating the Surrogate Modeling Capabilities of Continuous Time Echo State Networks

Continuous Time Echo State Networks (CTESNs) are a promising yet under-explored surrogate modeling technique for dynamical systems, particularly those governed by stiff Ordinary Differential Equations (ODEs). A key determinant of the generalization accuracy of a CTESN surrogate is the method of projecting the reservoir state to the output. This paper shows that of the two common projection methods (linear and nonlinear), the surrogates developed via the nonlinear projection consistently outperform those developed via the linear method. CTESN surrogates are developed for several challenging benchmark cases governed by stiff ODEs, and for each case, the performance of the linear and nonlinear projections is compared. The results of this paper demonstrate the applicability of CTESNs to a variety of problems while serving as a reference for important algorithmic and hyper-parameter choices for CTESNs.

On Accelerated Metaheuristic-Based Electromagnetic-Driven Design Optimization of Antenna Structures Using Response Features

Development of present-day antenna systems is an intricate and multi-step process requiring, among others, meticulous tuning of designable (mainly geometry) parameters. Concerning the latter, the most reliable approach is rigorous numerical optimization, which tends to be resource-intensive in terms of computing due to involving full-wave electromagnetic (EM) simulations. The cost-related issues are particularly pronounced whenever global optimization is necessary, typically carried out using nature-inspired algorithms. Although capable of escaping from local optima, population-based algorithms exhibit poor computational efficiency, to the extent of being hardly feasible when directly handling EM simulation models. A popular mitigation approach involves surrogate modeling techniques, facilitating the search process by replacing costly EM analyses with a fast metamodel. Yet, surrogate-assisted procedures feature complex implementations, and their range of applicability is limited in terms of design space dimensionality that can be efficiently handled. Rendering reliable surrogates is additionally encumbered by highly nonlinear antenna characteristics. This paper investigates potential benefits of employing problem-relevant knowledge in the form of response features into nature-inspired antenna optimization. As demonstrated in the recent literature, re-formulating the design task with the use of appropriately selected characteristic locations of the antenna responses permits flattening the functional landscape of the objective function, leading to faster convergence of optimization procedures. Here, we apply this concept to nature-inspired global optimization of multi-band antenna structures, and demonstrate its relevance, both in terms of accelerating the search process but also improving its reliability. The advantages of feature-based nature-inspired optimization are corroborated through comprehensive (based on three antenna structures) comparisons with a population-based search involving conventional (e.g., minimax) design problem formulation.

A novel cable-net buffer device for recovering reusable launch vehicle: Design, dynamic modeling, and performance analysis

At present, reusable launch vehicle technology (RLV) has become a research hotspot in the aerospace field and one of the major development trends in this industry. With reference to the aircraft carrier’s arresting cable technology, this paper puts forward a novel cable-net buffer device for recovering the first stage of launch vehicle. The device can transfer load through a cable-pulley mechanism, and then complete the first stage recovery by losing the kinetic energy and gravitational potential energy of the first stage through hydraulic dampers and preloaded counterweights. In order to analyze the performance of this cable-net buffer device, the finite element software ABAQUS is adopted to establish a rigid-flexible coupling dynamic model. In this model, the cable-pulley mechanism is simulated by connector and shell elements, the damping force of the hydraulic dampers by VUAMP subroutine, and the preloaded counterweights by mass point. And then, performance evaluation indexes of this buffer device are proposed in this paper. Meanwhile, the dynamic simulation, using the established dynamic model, is carried out under different device design parameters and working condition parameters. On the one hand, the simulation results show that the tower span and damping force have remarkable effects on the buffer performance. On the other hand, among all working condition parameters, only the fall point deviation has a significant effect on the buffer performance, but this effect is less than that of the design parameters. Finally, combing the dynamic model and the surrogate model technique, this paper obtains feasible boundaries of design parameters of the device for the subsequent detailed design of it. Researches in this paper will provide new ideas for the development of RLV.

Surrogate-based optimisation of process systems to recover resources from wastewater

Wastewater systems are transitioning towards integrative process systems to recover multiple resources whilst simultaneously satisfying regulations on final effluent quality. This work contributes to the literature by bringing a systems-thinking approach to resource recovery from wastewater, harnessing surrogate modelling and mathematical optimisation techniques to highlight holistic process systems. A surrogate-based process synthesis methodology was presented to harness high-fidelity data from black box process simulations, embedding first principles models, within a superstructure optimisation framework. Modelling tools were developed to facilitate tailored derivative-free optimisation solutions widely applicable to black box optimisation problems. The optimisation of a process system to recover energy and nutrients from a brewery wastewater reveals significant scope to reduce the environmental impacts of food and beverage production systems. Additionally, the application demonstrates the capabilities of the modelling methodology to highlight optimal processes to recover carbon, nitrogen, and phosphorous resources whilst also accounting for uncertainties inherent to wastewater systems.

Multi-fidelity Kriging extrapolation together with CFD for the design of the cross-section of a falling lifeboat

Surrogate modelling techniques such as Kriging are a popular means for cheaply emulating the response of expensive Computational Fluid Dynamics (CFD) simulations. These surrogate models are often used for exploring a parameterised design space and identifying optimal designs. Multi-fidelity Kriging extends the methodology to incorporate data of variable accuracy and costs to create a more effective surrogate. This work recognises that the grid convergence property of CFD solvers is currently an unused source of information and presents a novel method that, by leveraging the data structure implied by grid convergence, could further improve the performance of the surrogate model and the corresponding optimisation process. Grid convergence states that the simulation solution converges to the true simulation solution as the numerical grid is refined. The proposed method is tested with realistic multi-fidelity data acquired with CFD simulations. The performance of the surrogate model is comparable to an existing method, and likely more robust. More research is needed to explore the full potential of the proposed method. Code has been made available online at https://github.com/robertwenink/MFK-Extrapolation.

Active-learning-driven surrogate modeling for efficient simulation of parametric nonlinear systems

When repeated evaluations for varying parameter configurations of a high-fidelity physical model are required, surrogate modeling techniques based on model order reduction are desirable. In absence of the governing equations describing the dynamics, we need to construct the parametric reduced-order surrogate model in a non-intrusive fashion. In this setting, the usual residual-based error estimate for optimal parameter sampling associated with the reduced basis method is not directly available. Our work provides a non-intrusive error-estimator-based optimality criterion to efficiently populate the parameter snapshots, thereby, enabling us to effectively construct a parametric surrogate model. We consider parameter-specific proper orthogonal decomposition subspaces and propose an active-learning-driven surrogate model using kernel-based shallow neural networks (KSNNs), abbreviated as ActLearn-POD-KSNN surrogate model. The center location for each kernel, along with center-dependent kernel widths, can be learned for the KSNN by using an alternating dual-staged iterative training procedure. To demonstrate the efficiency of our proposed ideas, we present numerical experiments using four physical models, including incompressible Navier–Stokes equations. The ActLearn-POD-KSNN surrogate model efficiently predicts the solution at new parameter locations, even for a setting with multiple interacting shock profiles and a fluid flow scenario with Hopf bifurcation. We also provide an investigation of the surrogate’s performance when the available data is noisy.

An integrated surrogate model-driven and improved termite life cycle optimizer for damage identification in dams

Structural Damage Identification (SDI) is a crucial branch in the field of structural health monitoring, providing an essential support for the safe operation of structures. In this paper, a novel structural damage identification method based on a surrogate-assisted evolutionary optimization algorithm is proposed. This method incorporates cost-effective surrogate modelling techniques and swarm intelligence optimization algorithms with powerful global optimization capabilities. Three popular surrogate models are fused based on a weighted average strategy to construct an integrated surrogate model with stronger generalization ability and higher accuracy. In addition, the self-designed movement strategy is more suitable for the characteristics of Termite Life Cycle Optimizer (TLCO), allowing the improved TLCO to have stronger robustness and the capability to escape from local optimum. The effectiveness of the proposed method for solving the SDI problem is confirmed in a damaged dam model with different complexities. Some important findings are as follows: (i) Compared with the conventional SDI methods, which directly combines optimization algorithms and finite element models, the proposed method maintains reliable accuracy, while improving computational efficiency by a factor of more than 100. (ii) There is no direct relationship between the accuracy of the surrogate models and the effectiveness of the SDI method based on their combination with ITLCO. The proposed method can serve as a promising tool for damage identification in large-scale structures.

Mixed integer programming with kriging surrogate model technique for dispersion control of photonic crystal fibers

In this paper, mixed integer nonlinear programming (MINLP) optimization algorithm integrated with kriging surrogate-model is newly formulated to optimize the dispersion characteristics of photonic crystal fibers (PCFs). The MINLP is linked with full vectorial finite difference method (FVFDM) to optimize the modal properties of the PCFs. Through the optimization process, the design parameters can take real and/or integer values. The integer values can be used to selectively fill the PCF air holes to control its dispersion characteristics. However, the other optimization techniques deal with real design parameters where the PCF can be optimized using none or predefined infiltrated air holes. The MINLP algorithm is used to obtain an ultra-flat zero dispersion over a broadband of wavelength range from 1.25 to 1.6 μm using silica PCF selectively infiltrated with Ethanol material. To show the superiority of the proposed algorithm, nematic liquid crystal selectively infiltrated PCFs are also designed with high negative flat dispersion over wide range of wavelengths from 1.25 to 1.6 μm for the quasi transverse magnetic (TM) and the quasi transverse electric (TE) modes. Such designs have negative flat dispersions of − 163 ± 0.9 and − 170 ± 1.2 ps/Km nm, respectively over the studied wavelength range. Therefore, the MINLP algorithms could be used efficiently for the design and optimization of selectively filled photonic devices.

A Hybrid Data-Physics Framework for Reservoir Performance Prediction with Application to H2S Production

Summary Model-based reservoir management workflows rely on the ability to generate predictions for large numbers of model and decision scenarios. When suitable simulators or models are not available or cannot be evaluated in a sufficiently short time frame, surrogate modeling techniques can be used instead. In the first part of this paper, we describe extensions of a recently developed open-source framework for creating and training flow network surrogate models, called FlowNet. In particular, we discuss functionality to reproduce historical well rates for wells with arbitrary trajectories, multiple perforated sections, and changing well type or injection phase, as one may encounter in large and complex fields with a long history. Furthermore, we discuss strategies for the placement of additional network nodes in the presence of flow barriers. Despite their flexibility and speed, the applicability of flow network models is limited to phenomena that can be simulated with available numerical simulators. Prediction of poorly understood physics, such as reservoir souring, may require a more data-driven approach. We discuss an extension of the FlowNet framework with a machine learning (ML) proxy for the purpose of generating predictions of H2S production rates. The combined data-physics proxy is trained on historical liquid volume rates, seawater fractions, and H2S production data from a real North Sea oil and gas field, and is then used to generate predictions of H2S production. Several experiments are presented in which the data source, data type, and length of the history are varied. Results indicate that, given a sufficient number of training data, FlowNet is able to produce reliable predictions of conventional oilfield quantities. An experiment performed with the ML proxy suggests that, at least for some production wells, useful predictions of H2S production can be obtained much faster and at much lower computational cost and complexity than would be possible with high-fidelity models. Finally, we discuss some of the current limitations of the approach and options to address them.

Proper Orthogonal Decomposition-Based Surrogate Modeling Approximation for Aeroengines Nonlinear Unbalance Responses

Abstract Finite element (FE) models to simulate aeroengines rotordynamics become more and more complex thanks to the improvement of computational resources. Numerical simulations for industrial scale applications are nevertheless expensive in terms of computational effort because of their size and the presence of nonlinearities. In many-query frameworks (sensitivity or reliability analysis, model calibration, etc.), parametric studies have to be performed and the FE models are generally simplified or strongly reduced with a consequent loss of accuracy. Recent machine learning methods can be employed to build faster-to-evaluate surrogates of the original model. Nevertheless, this type of offline approach can be very demanding in terms of number of full order evaluations to reach a sufficient accuracy. The challenge is therefore to control the burden associated with the evaluation of the design of experiments (DoE) and the surrogate model training phase. This work presents a surrogate modeling technique to quickly and accurately reproduce the nonlinear unbalance responses of industrial scale rotor-dynamical models submitted to rotor–stator contact. About 20DOFs of the industrial FEM are considered in this work and a proper orthogonal decomposition (POD)-based approximation (surrogate modeling technique known as POD-SM or nonintrusive POD in the literature) of their frequency dependent unbalance response are presented. In order to analyze the modeling uncertainties of a specific shaft's support of a modern aeroengine, its dynamical parameters (stiffness, damping, nonlinear contact parameters) are studied within a specific variation range, defining the design space covered by this study.

Surrogate modeling in irreversible electroporation towards real-time treatment planning

In this paper, we develop surrogate models that can replace expensive predictive models and account for uncertainties in real-time treatment planning for irreversible electroporation of liver tumors. Standard non-intrusive surrogate modeling techniques that account for the model uncertainty and reduce the computational cost, such as polynomial chaos expansion and Gaussian process regression with conventional kernels, often do not capture the true physical behavior of the treatment outcome as required in the context of treatment planning. We improve the Gaussian process regression model by modifying the kernel function to a non-stationary Gibbs kernel with a support vector machine-based classifier in its length scale definition. This proposed model is compared with the standard surrogates in terms of their performance and accuracy. Our model is able to accurately replicate the behavior of the biophysics-based predictive model. There is a decrease of at least 81% in the overall root-mean-square error for treatment outcome when compared to the Gaussian process regression model with conventional kernels. Furthermore, we illustrate the application of the proposed surrogate model in treatment planning to address a voltage optimization problem for complete tumor ablation. Surrogate-assisted treatment planning exhibited good performance while maintaining similar levels of accuracy in comparison to treatment planning based on biophysical models. Finally, the effect of uncertainty in tissue electrical conductivities on the optimal voltage value is discussed.