Estimation of temperature-dependent thermal conductivity using proper generalised decomposition for building energy management

Abstract

A proper generalised decomposition for solving inverse heat conduction problems is proposed in this article as an innovative method offering important numerical savings. It is based on the solution of a parametric problem, considering the unknown parameter as a coordinate of the problem. Then, considering this solution, all sets of cost function can be computed as a function of the unknown parameter of the defined domain, identifying the argument that minimises the cost function. In order to illustrate the applicability, the method is used to solve a non-linear inverse heat conduction problem to determine a temperature-dependent thermal conductivity. Then, a comparison is carried out with the local sensitivity and the genetic algorithm methods. It is shown that the proper generalised decomposition method estimates the unknown parameter with the same accuracy as the other two methods. Due to its advantage in terms of reducing the complexity, the method was then used to solve a transient three-dimensional nonlinear heat transfer inverse problem. The results have shown that the method is appropriate to determine the unknown parameter with a low computational cost. Downloaded from Furthermore, the main advantage of the technique is its low capacity for storage, which can be used, as an inverse method, for building energy management and extended to evaluate thermal bridges from on-site measurements.

On the optimal experiment design for heat and moisture parameter estimation

“…A normal distribution with a standard deviation of σ = 0.01 was assumed. The inverse problem is solved using Levenberg-Marquardt method [6,21,33]. After the solution of the N e × 30 inverse problems, the empirical mean square error is computed: where P • is the estimated parameter by the resolution of the inverse problem.…”

Multi-level Convolutional Autoencoder Networks for Parametric Prediction of Spatio-temporal Dynamics

“…. , 2 10 , 2 11 . Two dense layers are appended after the TCN block, and the encoded latent dimension is n l = 100. l 2 regularization with a penalty factor λ = 1e−8 is used in the loss function.…”

“…The governing equations of the full order model are then projected on to the lower dimensional subspace by choosing an appropriate test basis. Other linear basis construction methods include balanced truncation [4,5], reduced basis methods [6,7,8], rational interpolation [9], and proper generalized decomposition [10,11]. Linear basis ROMs have achieved considerable success in complex problems such as turbulent flows [12,13,14] and combustion instabilities [15,16].…”

Transport Map sampling with PGD model reduction for fast dynamical Bayesian data assimilation

“…23,24 The construction of the PGD solution is made in an offline phase and leads to a cost-effective evaluation of the numerical model depending on parameters in the online inversion phase. 25,26 The issue of the model evaluation cost is emphasized in Bayesian inference where the posterior density has to be explored to derive useful information on parameter pdfs such as mean, standard deviation, maximum, and marginals. Those quantities require the computation of multidimensional integrals over the parametric space.…”