It has become a major challenge for the colocation technology to schedule jobs appropriately and accurately and minimize the impact on online jobs in colocation scenarios. To address these limitations, we propose a scheduling strategy driven by a multi-dimensional resource mixing interference model, which uses deep neural networks to build an accurate performance interference model by monitoring multi-dimensional resource metrics, and employs a pre-run initial scheduling, run-time dynamic resource adjustment, and run-time re-migration strategy to ensure that the performance of online jobs in colocation scenarios is not affected as much as possible.