Use of Static Surrogates in Hyperparameter Optimization

Lakhmiri, Dounia; Digabel, Sébastien Le

doi:10.1007/s43069-022-00128-w

Cited by 1 publication

(1 citation statement)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The problem mentioned above can be handled by adjusting the hyper-parameters of the neural network intelligently, including the static and dynamic optimization neural network (SONN and DONN). − The first one adjusts the hyper-parameters outside the iterative process of the neural network; the second dynamically adjusts the hyper-parameters inside the iterative process. , The typical static optimization techniques are grid search and evolution strategies. − These techniques can explore different neural network hyper-parameters, but slowly. On the other hand, gradient methods and neural network structure adjustment methods are two categories of dynamic optimization techniques .…”

Section: Introductionmentioning

confidence: 99%

Intelligent Dynamic Prediction Model of Pump Pressure Lift for a Petroleum Pipeline System with Nonisothermal Batch Transportation

Jiang,

Yuan,

Chen

et al. 2023

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

Accurate prediction of pump characteristics is pivotal for optimizing transportation schemes and device operations in long-distance petroleum pipelines. In response to the variable nature of pump characteristics due to production circumstances, a novel dynamic prediction model of pump pressure lift is proposed by coupling the neural network, the intelligent evolutionary strategy, and an adaptive method of a neural network coefficient matrix. This model enables parameter adjustments to align with optimal directions through intelligent optimization strategies in which the differential evolution performs better than the genetic algorithm and particle swarm optimization. The proposed model exhibits superior accuracy and computational efficiency compared with classical static and dynamic prediction models. Notably, the model achieves a prediction error reduction of 1.34%, surpassing existing dynamic models while offering an 87.93% computation time reduction when contrasted with the static model

show abstract