Testing resource allocation is a core problem in software testing. Most of the existing related studies assume that the software architecture is static and rarely consider cost constraints. To address this problem， a software testing resource allocation algorithm for dynamic changes in architecture was proposed. Firstly， a multi-stage multi-objective multi-constraint testing resource allocation model with dynamically changing architecture was constructed. Then， based on parameter re-estimation and generalized differential evolution， the population re-initialization was added to the algorithm， which was able to reduce the algorithm search space and improve the algorithm performance. Finally， a new repair processing mechanism was added to the algorithm， which was able to eliminate the invalid solutions generated by the algorithm effectively. Compared with the solution sets obtained by the Multi-Objective Differential Evolution based on Weighted Normalized Sum （WNS-MODE） algorithm and Dynamic Testing Resource Allocation based on Generalized Differential Evolution 3 （DTRA-GDE3） algorithm， the solution set obtained by the proposed algorithm has the capacity value improved by about 11.81 times and 0.39 times respectively. In terms of coverage value metrics， the proposed algorithm completely covered the WNS-MODE algorithm and improved 81 percentage points with respect to the DTRA-GDE3 algorithm. In terms of the super volume value metrics， the proposed algorithm improved nearly 6 and 9 times， respectively. Experimental results show that the proposed algorithm can better adapt to the dynamic changes in software architecture， can provide more and better testing resource allocation schemes for dynamic testing of software products， and meets the dynamic changes in user requirements.