To address the issues such as slow convergence， insufficient convergence accuracy， and the inability to jump out of the local optimum in Harris Hawks Optimization （HHO） algorithm， an improved algorithm based on Refracted Opposition-Based Learning （ROBL） and adaptive strategy was proposed. By introducing a refracted opposition-based learning strategy， opposition solutions were generated to broaden the search scope， thereby enhancing both the convergence speed and the global exploration capability of the algorithm. Concurrently， adaptive inertia weights and nonlinear energy decay factors were employed to adjust the exploration and exploitation capabilities of the algorithm dynamically. Besides， an improved adaptive t distribution variation was incorporated to mutate the optimal positions， thereby enhancing the algorithm's capacity to jump out of the local optimum. While preserving the population diversity， the improved algorithm accelerated convergence， enhanced global search capability and convergence accuracy. In comparative experiments conducted on 12 benchmark functions， it is validated that compared to swarm intelligence algorithms， the proposed HHO algorithm has the highest convergence accuracy on all the functions. Furthermore， in the benchmark test function experiments， the effectiveness of single improvement strategies was verified， as well as the superiority of employing combinations of multiple strategies over using single strategies alone.