To address the performance and technical bottlenecks of a particle swarm optimization algorithm in tackling optimization problems of high-dimensional, large-scale, multivariate coupling, multi-modal, multi-extreme attribute vulnerable to premature convergence, a multi-operator selection and fusion mechanism with biased roulette is established based on the behavioral learning operator of particle swarm optimization and three differential mutation operators with different learning preferences and a multi-operator collaborative particle swarm optimization algorithm with biased roulette is proposed(MOCPSO). For balancing the exploration-exploitation trade-off, MOCPSO first groups the iterative swarm into several subswarms with different learning tasks according to the fitness, where each subswarm configures a differential mutation operator that the differential mutation vectors selected among the exemplars of all subswarms through roulette selection, to pre-learn and optimize the iterative swarm and their exemplar particles. Then all subswarms are merged undergoing behavioral learning operation of the particle swarm optimization to improve the global convergence. Finally, for guaranteeing the diversity of algorithm convergence, the MOCPSO incorporates an elitist learning strategy to guide iterative swarm escaping the possible local traps by performing Gaussian perturbation on the current global best. Experimental results show that the proposed MOCPSO algorithm possesses stronger and more competitive optimization properties than five state-of-the-art swarm intelligence algorithms in solving the CEC2014 benchmark test suit.