There is only one guiding solution in the search equation of Gaussian bare-bones artificial bee colony algorithm (ABC-BB), which is easy to result in the problem of premature convergence and trapping into the local minimum. In order to enhance the capability of escaping from local minimum without loss of the exploitation ability of ABC-BB, a new triangle search strategy is proposed. The candidate solution is generated among the triangle area formed by current solution, global best solution and any randomly selected elite solution to avoid the premature convergence problem. Moreover, the probability of crossover is controlled dynamically according to the successful search experience, which can enable ABC-BB to adapt all kinds of optimization problems with different landscapes. The experimental results on a set of 23 benchmark functions and two classic real-world engineering optimization problems show that the proposed algorithm is significantly better than ABC-BB as well as several recently-developed state-of-the-art evolution algorithms.