Harmony search (HS) is a form of stochastic meta-heuristic inspired by the improvisation process of musicians. In this study, a modified HS with a hybrid cuckoo search (CS) operator, HS-CS, is proposed to enhance global search ability while avoiding falling into local optima. First, the randomness of the HS pitch disturbance adjusting method is analyzed to generate an adaptive inertia weight according to the quality of solutions in the harmony memory and to reconstruct the fine-tuning bandwidth optimization. This is to improve the efficiency and accuracy of HS algorithm optimization. Second, the CS operator is introduced to expand the scope of the solution space and improve the density of the population, which can quickly jump out of the local optimum in the randomly generated harmony and update stage. Finally, a dynamic parameter adjustment mechanism is set to improve the efficiency of optimization. Three theorems are proved to reveal HS-CS as a meta-heuristic algorithm. In addition, 12 benchmark functions are selected for the optimization solution to verify the performance of HS-CS. The analysis shows that HS-CS is significantly better than other algorithms in optimizing high-dimensional problems with strong robustness, high convergence speed, and high convergence accuracy. For further verification, HS-CS is used to optimize the back propagation neural network (BPNN) to extract weighted fuzzy production rules. Simulation results show that the BPNN optimized by HS-CS can obtain higher classification accuracy of weighted fuzzy production rules. Therefore, the proposed HS-CS is proved to be effective.