Efficient Parallel Computing Of Inverse Heat Conduction Problem Without Heat Source

With more and more demand in engineering field, and comprehenseive researches in mathematics methods and tools, the inverse problem, which gets reason from result and gains nature from phenomenon, is applied in many engineering fields. As the inverse problem's background is very extensive and important, the problem's research and solution require many cross subject, so the theory of its research is still new and challenging.Inverse heat conduction problem(IHCP) is a classical Math-matical Physics inverse problem. The IHCP studied in the thesis is to use the temperature measured somewhere within the boundaries to derive the temperature function.First of all, the thesis describes 2-D IHCP's math model and 2-D IHCP's math model. Solving 2-D IHCP depends on its forward problem, so a detailed study of the 2-D forward problem solving is adressed. In the thesis, the particle swarm optimization (PSO) algorithm is used to solve IHCP faster and more accurate. Using PSO algorithm to solve three different meshing 2-D IHCP, we can conclude that the standard PSO algorithm is not suitable for solving fine-mesh 2-D IHCP.Anglysing the experimental data, the thesis identifies that there are two main reasons for poor accuracy of PSO algorithm solving fine-mesh 2-D IHCP:Fine mesh division leads the particle having too many dimensions, makes the search process is very easy to fall into local optimum. And too many dimensions in a particle makes some dimensions'accuracy is continually reducing with the search going.Considering the particular application fine-mesh 2-D IHCP, the thesis gives an improved strategy for the standard PSO algorithm.Results show that the improved PSO algorithm has a great improve in global search capability and solution precision.As the calculation for solving fine-mesh 2-D IHCP is very large, the thesis implements the parallel improved PSO algorithm to accelerate the calculation speed. In the parallel laboratory, a parallel computer cluster system is built on the MPI platform, and the parallel algorithm performances are measured by the specific test.