面向银行业务的交易量预测与告警研究

银行一般都有多种交易系统并存，当这些分散的交易系统出现故障时，运维人员难以从海量的日志中定位故障。针对以上问题，使用SparkStreaming、Spark ML、Hadoop、ELK等技术，基于决策树回归模型，设计并实现了一个面向银行业务的交易量预测与告警平台。该平台能够实时监控各交易系统近期交易量，并对各个交易系统不同时段的交易量进行预测，预测值作为交易系统交易量的动态阈值，平台能够根据阈值对异常的系统进行实时告警。真实环境下的运行结果表明，平台很好地满足了银行交易系统运维的需求。

Research on Prediction and Alarm of Transaction Volume Oriented to Banking Business

Banks generally have a wide variety of co-existing transaction systems, when these decentralized transaction systems malfunction, it is difficult for the maintenance support engineer to locate the cause of malfunction from the massive logs. For the problems above, Spark Streaming, Spark ML, ELK and other technologies are used to design and implement the monitoring and alarming platform oriented to banking business, which is based on decision tree regression model. The platform can monitor the transaction volume of the transaction system in real time and predict the transaction volume of different transaction systems in different periods. The predicted values are dynamic thresholds of the transaction volume. The platform can alarm the abnormal system according to the threshold in real time. Operation results of real environment show that the platform satisfies the demand of maintenance of the bank transaction systems well.