Abstract: Underprediction will lead to the decrease of service capability, the increase of request violation rate and rejection rate, and finally the degradation of service quality. In the current research on load prediction strategy of container cloud, there is no effective solution to the problem that network application access request is lost due to prediction failure. A underprediction coping strategy (PFCS) was proposed, and an emergency scaling algorithm (ESA) was proposed based on the PFCS. This strategy can effectively detect the occurrence of prediction failure, activate the emergency prediction model in time, and provide another option for scaling decisionl; This strategy can shorten the failure time of horizontal scaling policy to deal with burst traffic, reduce the request rejection rate, and improve the overall quality of service. The experiment uses real network application traffic data to verify the effectiveness of ESA in improving quality of service. The comparison experiment is used to evaluate the response effect of PFCS on prediction failure. The experimental results show that compared with the scaling strategy using only prediction algorithm, udder the condition of not affecting the prediction accuracy, PFCS can make rejection rates decrease by 19.8%-23.0%on average. The experimental results demonstrate the effectiveness of ESA and PFCS.