Effectiveness of an estimation strategy for radiation source distribution using generalized Tikhonov regularization

一般化チホノフ正則化を用いた放射線源分布推定法の有効性

山田 進  ; 町田 昌彦  

Yamada, Susumu; Machida, Masahiko

It is essential to detect the distribution of the radiation source for safely proceeding with the decommissioning of reactor buildings. It has been reported that the source distribution can be estimated using a model constructed with uniform cells based on measured air dose rates by minimizing an evaluation function with LASSO (Least Absolute Shrinkage and Selection Operator) regression. Moreover, when the cells are non-uniform, the distribution can be estimated using fused LASSO, which minimizes the evaluation function that incorporates the connectivity between adjacent cells. However, since LASSO regression tends to shrink parameters with less influence toward zero, the estimated source distribution may become discrete, even when the actual distribution extends across multiple cells. This issue may persist even using fused LASSO. Therefore, to address the issue, we focus on generalized Tikhonov regularization. This is a method to find the vector minimizing a function composed of the residual sum of squares plus an L2-regularization term. When we apply this regularization for estimating the source distribution based on measured air dose rates, we can control the connectivity between adjacent cells by varying the regularization term. In this paper, we evaluate the effectiveness of generalized Tikhonov regularization in estimating source distributions for models with widespread sources, and we demonstrate that this method can accurately estimate such distributions. As a result, we confirm that the source distribution can be estimated with high accuracy using generalized Tikhonov regularization with a second-order differential operator, which promotes a constant rate of change in concentration across multiple cells.