The prediction of minimum-energy protein structures starting from a sequence of amino acids is a computationally challenging problem even in simplified lattice protein models. A hybrid evolutionary model is designed and tested in the current paper to address this well-known NP-hard problem. Hill-climbing strategies are integrated in the search operators and a meaningful diversification of genetic material occurs during the population evolution. The main features of the proposed algorithm refer to a weak hill-climbing application of uniform crossover and pull move transformations and the randomization of genetic material based on the fingerprint of the protein conformations. Numerical experiments are performed for several difficult bidimensional instances from lattice models (the hydrophobic-polar model and functional model proteins). The results are competitive with those obtained by related population-based optimization algorithms.