於前人研究可得知，相同正算理論程式若採用不同回算演繹技巧將會影響回算最終結果之呈現，且傳統數值分析方法(如：迭代法)常會遭遇到回算起始模數值選用與陷入局部極小值問題造成回算誤差；而新型啟發式演算法(如：基因演算法) 雖然全域搜尋能力佳，但在精度要求下常造成回算時間過長等缺點。因此，本研究除探討回顧國內外鋪面撓度回算程式所採用的回算方法及數值分析技巧外，擬針對落重撓度試驗動力特性，使用全域搜尋之基因演算法與局部搜尋之二分法並稱為混合式基因演算法(Hybrid Genetic Algorithm, 簡稱HGA)，且以此方法為回算工具進而發展成動力回算程式DBFWD-HGA。
The pavement FWD device, which can effectively know mechanics of the pavement well and provide valuable information of pavement management and maintenance, is a major equipment used to evaluate the deflections of pavements around the world. Moreover, the test measurement data is comprehensively analyzed by using the backcalculation. As the design method of pavement evolves from traditional method into mechanics-empirical design method, the backcalculation analysis turns out to become very crucial.
According to former research, same forward calculation program with different methods of backcalculation will lead to different results. Furthermore, traditional numerical methods such as iterative method often encounter backcalculation errors due to the initial value and local minimum value. Additionally, although the new-type heuristic algorithm such as genetic algorithm is good at global search, it still has the flaw of long backcalculation time with the requirement of resolution. Hence, this research not only explores the backcalculation and numerical methods introduced by backcalculation program around the world but also capitalizes on global search of genetic algorithm associated with local search of bisection method to form a Hybrid Genetic Algorithm(HGA). Accordingly, dynamic backcalculation program DBFWD-HGA can be formed based on HGA.
In order to explore and compare same forward calculation program with respect to different algorithms, this research simulates the results of flexible and rigid pavement and field case study by using DBFWD, DBFWD-GA, and DBFWD-HGA. In the theoretical backcalculations, the results show that using the local search method can not only increase the search capability and convergence rate of genetic algorithm but also effectively solve the fine-tuning deficiency of the algorithm and save operation time. Therefore, in the means of searching the optimum solution, this method can effectively improve the operation efficiency of genetic algorithm. However, regarding the case analysis, the results show that the method is not superior to traditional DBFWD and MODULUS due to limited variable study on factors affecting the backcalculations. Besides, calculation time consumed by genetic algorithm is much longer than iterative method. Though the calculation time of DBFWD-HGA can be lowered by 0.27~0.75 times (27%~75%), it is still more than 5 times that of iterative method. In this case, this method is not yet efficient at the being.