We developed a traffic prediction system which enhances a traffic information service. The prediction method is based on time series analysis and is applicable to short to long term prediction. Traffic information system are real-time and real-world system therefore it suffers various kind of disturbance from environment. To preserve traffic prediction quality, we need fundamental treatment on overall system so that the prediction engine be tolerant toward incomplete traffic data feed or non-stationary traffic data. A solution for incomplete data feed is a combination of data for multiple links. A solution for non-stationary traffic is a traffic simulation dedicated to traffic accidents. With these enhancements toward cyber disturbance and physical disturbance, the system resiliency can be higher.