To realize early prediction of spontaneous combustion of overlong double coal roadway, this paper proposes a CO trend analysis method based on the big data of the underground monitoring system when the spontaneous combustion mark-gas analysis method is invalid. Meanwhile, to screen out and remove fluctuation disturbance of CO caused by underground diesel vehicle exhaust, the concept of background CO volume fraction which can only reflect the relatively slow oxidation of coal is put forward. The background CO volume fraction screening method is established on the basis that for an independent ventilated investigation area of a mine, there is always an extreme moment that no diesel vehicle works (or exhaust gas does not interfere all diesel car) in a long enough time, and the CO volume fraction value is recorded by the monitoring system. Through the comparison of the screening results of a 0.125-day time unit cycle and a 0.5-day time unit cycle, it is found that as the time unit increases, the CO volume fraction curve becomes lower, the fluctuation decreases, and the background CO volume fraction curve becomes clearer, suggesting the objective existence of background CO volume fraction. Taking a part of the shortest spontaneous combustion period as the examination period, the spontaneous combustion of coal pillars is predicted based on the trend development of the background CO volume fraction. Combined with the experience of Hongqingliang coal mine, the CO increasing trend(increasing rate k₁^*=0.607×10^-6 d^-1) is obtained as the critical index of spontaneous combustion risk prejudgment. The spontaneous combustion risk warning is divided into three levels. When k₁≥k₁^*, it is first-level warning, which triggers key inspections, and upgrades the daily comprehensive inspection to the targeted inspection with more manpower and material resources. When k₁1<k₁^*, it is second-level alert, which requires to find out the causes. When k₁≤0, there is no spontaneous combustion risk. The predicted results meet the engineering requirements.