微型电脑应用

为了在电力行业数据中心机房预警系统中何融入机器学习功能，解决目前设备预警系统历史数据利用率低、预警信息需要人工判断、无法利用机器学习模型等问题，以BP神经网络、Stacking模型及Hadoop集群技术为基础，设计一款运维改进方案。探讨系统功能、预警模块、系统管理模块、模型的重复使用及训练的方案。通过应用大数据进行测试，结果表明传统模型的训练需要的时间较长。对于单个模型而言，其训练时长会小于采用Stacking集成模型进行训练所需的平均时长。由此可知，不论是单个模型还是Stacking模型，其模型预测方面的耗时几乎可以实现实时预测。

The current work aims to study to integrate the machine learning function into the early warning system of the data center computer room in the power industry, improve the utilization rate of historical data, the manual judgment for early warning information. In the old system, the machine learning model can not be used. Based on backpropagation neural network, Stacking model and Hadoop cluster technology, a scheme including the reuse and training of system functions, early warning modules, system management modules and models is designed. The system performance is tested by big data, and the results show that the training time of the traditional model is longer. For a single model, the training time will be less than the average time required for training with the Stacking ensemble model. Whether it is a single model or a Stacking model, the time of the model prediction can almost achieve real-time prediction.