Genetic algorithms (GAs) are the adaptation methods broadly applicable to many classes of problems. Adaptation to changing environments is one of the important classes of such problems. Continuous search for the solutions by the GA is the fundamental mechanism for adaptation, and therefore to avoid convergence by maintaining the diversity is an intrinsic requirement for successful search. The authors have proposed to utilize the thermodynamical genetic algorithms (TDGA), a genetic algorithm which maintains the diversity of the population by evaluating its entropy, for the problem of adaptation to changing environments. However, if the environmental change has a recurrent nature, a memory-based approach, i.e., to memorize the results of past adaptation and to retrieve them as candidates for the solution, will be a smart strategy. In the present paper, the authors combine the memory-based approach with TDGA as an adaptation algorithm to changing environments. The adaptation ability of the proposed method is verified by computer simulations taking recurrently varying knapsack problems as examples.