The identification method based on the traditional Proper Orthogonal Decomposition (POD) reduced-order model has the problem of low efficiency, due to the large amount of both data and computation, when dealing with complicated problems with a large number of spatially distributed nodes. To deal with this issue, an improved POD reduced-order model is proposed in this work. The improved POD reduced-order model only requires the establishment of a three-dimensional (3D) database of training samples varied with both time and measurement locations. Therefore, the amount of data and computation is independent of the total number of spatially distributed nodes, which enables the amount of data and computation to be greatly reduced. To identify thermal parameters in heat conduction problems, a database of transient temperature field is constructed with different training parameters and space nodes by using polygonal boundary element method, and a set of POD basis vectors is obtained by the POD reduced-order model. Then, a surrogate model combined with the improved particle swarm optimization (PSO) is employed for the identification of thermal parameters. Three different inverse heat conduction problems are designed and analyzed to verify the performance of the improved methodology. The results show that the efficiency of the modified method is superior to the traditional POD method with comparable accuracy. The more of the number of spatially distributed nodes, the more obvious advantages of the efficiency. Furthermore, this method has been tested on noisy data, proving its reliability in dealing with problems arising from measurement errors.
Read full abstract