In order to effectively solve dynamic optimization problems, a new hybrid memory scheme that consists of short-term memory and long-term memory is proposed. Information to be memorized includes the best individual and the probability vector of current population. Information of short-term memory is extracted to build the next population in each generation. Long-term memory is assigned for the short-term memory when a environmental change is detected. A new genetic algorithm is thus constructed based on the hybrid memory. Performance of the algorithm is verified in different environments including non-cyclic, cyclic, and cyclic with noise. Computation results indicate that this algorithm is superior to similar algorithms in dealing with dynamic optimization problems.