基于Spark Streaming流回归的煤矿瓦斯浓度实时预测

Real-time prediction of gas concentration in coal mine based on Spark Streaming Linear Regression

为了实时分析瓦斯监测流数据并对瓦斯浓度进行准确预测以实现瓦斯灾害实时预警,以实时流数据处理框架Spark Streaming构建基于流回归的瓦斯浓度实时预测系统。系统采用分布式流处理技术,可使基于回归算法的瓦斯浓度预测模型更新周期达到秒级,提高了瓦斯浓度预测精度,满足流式大数据处理的实时性要求。实验表明:应用Spark Streaming流回归预测系统在采样周期为5s的瓦斯监测数据流上进行实时预测时,预测平均均方根误差随模型更新周期的缩短而减小,模型更新周期可达15s,且更新周期为45s时预测总均方根误差最小,既能保证预测精度,又能提高瓦斯灾害预警时效。

In order to analyze the streaming data of gas monitoring in real - time and predict the gas concentration accurately, so as to achieve the real - time warning of gas disaster, a real - time prediction system of gas concentration based oil Stream- ing Linear Regression was constructed by using the real -time streaming data processing framework Spark Streaming. The system adopted the distributed stream processing technology, which could make the update cycle of the gas concentration pre- diction model based on regression algorithm reach the second level, improve the prediction accuracy of gas concentration, and meet the real - time requirement of streaming large data processing. The experiments showed that when applying the predic- ting system based on Spark Streaming Linear Regression to carry out the real - time prediction on the gas monitoring data stream with the sampling cycle of 5 seconds, the average root -mean -square error （RMSE） of prediction decreased with the shortening update cycle of the model. The update cycle of the model could reach 15 seconds, and the total RMSE was the smallest when the update cycle was 45 seconds, which can ensure the accuracy of prediction, and improve the warning timeli- ness of gas disaster.