Nicolai Palm

Mr. Nicolai Palm


Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 October 2022 in Sustainability
Verified
Journal Article
Open access
PDF file
NPNicolai Palm
Nicolai Palm
MLMarkus Landerer
Markus Landerer
HPHerbert Palm
Herbert Palm
Within a disruptively changing environment, design of power systems becomes a complex task. Meeting multi-criteria requirements with increasing degrees of freedom in design and simultaneously decreasing technical expertise strengthens the need for multi-objective optimization (MOO) making use of algorithms and virtual prototyping. In this context, we present Gaussian Process Regression based Multi-Objective Bayesian Optimization (GPR-MOBO) with special emphasis on its profound theoretical background. A detailed mathematical framework is provided to derive a GPR-MOBO computer implementable algorithm. We quantify GPR-MOBO effectiveness and efficiency by hypervolume and the number of required computationally expensive simulations to identify Pareto-optimal design solutions, respectively. For validation purposes, we benchmark our GPR-MOBO implementation based on a mathematical test function with analytically known Pareto front and compare results to those of well-known algorithms NSGA-II and pure Latin Hyper Cube Sampling. To rule out effects of randomness, we include statistical evaluations. GPR-MOBO turnes out as an effective and efficient approach with superior character versus state-of-the art approaches and increasing value-add when simulations are computationally expensive and the number of design degrees of freedom is high. Finally, we provide an example of GPR-MOBO based power system design and optimization that demonstrates both the methodology itself and its performance benefits.

21 October 2022 in IEEE Access
Verified
Journal Article
Open access
NPNicolai Palm
Nicolai Palm
FSFlorian Stroebl
Florian Stroebl
HPHerbert Palm
Herbert Palm
Parametric models allow to reflect system behavior in general and characterize individual system instances by specific parameter values. For a variety of scientific disciplines, model calibration by parameter quantification is therefore of central importance. As the time and cost of calibration experiments increases, the question of how to determine parameter values of required quality with a minimum number of experiments comes to the fore. In this paper, a methodology is introduced allowing to quantify and optimize achievable parameter extraction quality based on an experimental plan including a process and methods how to adapt the experimental plan for improved estimation of individually selectable parameters. The resulting parameter-individual optimal design of experiments (pi-OED) enables experimenters to extract a maximum of parameter-specific information from a given number of experiments. We demonstrate how to minimize variance or covariances of individually selectable parameter estimators by model-based calculation of the experimental designs. Using the Fisher Information Matrix in combination with the Cramer-Raó inequality, the pi-OED plan is reduced to a global optimization problem. The pi-OED workflow is demonstrated using computer experiments to calibrate a model describing calendrical aging of lithium-ion battery cells. Applying bootstrapping methods allows to also quantify parameter estimation distributions for further benchmarking. Comparing pi-OED based computer experimental results with those based on state-of-the-art designs of experiments, reveals its efficiency improvement. All computer experimental results are gained in Python and may be reproduced using a provided Jupyter Notebook along with the source code. Both are available under https://github.com/nicolaipalm/oed .

9 July 2025 in Journal of Mathematical Sciences
Verified
Journal Article
Open access
PDF file
NPNicolai Palm
Nicolai Palm
HPHerbert Palm
Herbert Palm
Multi-criteria decision-making requires knowledge about trade-offs between conflicting targets. Multi-objective optimization (MOO) strives to provide this information in terms of the set of Pareto optimal solutions. From an application point of view, a taxonomic scheme for selecting or constructing appropriate MOO algorithms under the constraint of individually selectable solution attributes for a given black box problem is of outstanding benefit. In this context, we introduce a mathematical framework to construct MOO algorithms tailored to individual needs. The approach is based on the composition of black-box functions with tailor-made support functions. The composition must exhibit a basic property, which we call Pareto reflective, to preserve Pareto points when support functions are concatenated with a black-box function. Mathematically, we prove some of the support functions’ fundamental structures and, thereby, class inducing invariance properties. This theoretical base, in turn, allows to derive a rigid set of construction rules such as edge point identification or equidistant grid scanning of the Pareto front. The related methodology bears two major advantages: it allows to classify and tailor MOO algorithms according to desired Pareto front search attributes, and secondly, it enables to significantly expand a MOO algorithm’s applicability area.

26 February 2024 in arXiv
Verified
Preprint
Open access
PDF file
NPNicolai Palm
Nicolai Palm
TNThomas Nagler
Thomas Nagler
Resampling methods such as the bootstrap have proven invaluable in the field of machine learning. However, the applicability of traditional bootstrap methods is limited when dealing with large streams of dependent data, such as time series or spatially correlated observations. In this paper, we propose a novel bootstrap method that is designed to account for data dependencies and can be executed online, making it particularly suitable for real-time applications. This method is based on an autoregressive sequence of increasingly dependent resampling weights. We prove the theoretical validity of the proposed bootstrap scheme under general conditions. We demonstrate the effectiveness of our approach through extensive simulations and show that it provides reliable uncertainty quantification even in the presence of complex data dependencies. Our work bridges the gap between classical resampling techniques and the demands of modern data analysis, providing a valuable tool for researchers and practitioners in dynamic, data-rich environments.