基于LSTM模型的工作日期间机房温度预测方法

Computer Room Temperature Prediction Method Based on LSTM Model

数据中心机房大多采用不间断空调制冷来对机房的温度进行控制,这样不仅能耗高且效率低,因此如果能对机房温度进行准确预测,并根据预测结果合理控制空调可以为企业节省巨大成本。针对这一情况,文章提出了一种工作日期间机房温度预测方法,只考虑工作日期间采集的机房温度数据,在相邻的两段工作日之间,利用数据拟合算法对首尾数据进行拟合,对拼接处的部分数据用拟合数据代替,然后利用多层长短期记忆神经网络(LSTM)模型对处理后的数据和原始数据分别进行预测,比较两者的预测结果,最终证明用本文提出的方法对数据处理后可以得到更精确的预测结果。

Most data center computer rooms use uninterrupted air conditioning to control the temperature of the computer room.This not only has high energy consumption and low efficiency,so if the temperature of the computer room can be accurately predicted and the air conditioning can be properly controlled based on the prediction results,it can save huge profits for the enterprise cost.In view of this situation,a method for predicting the temperature of the computer room during the working day is proposed.Only the temperature data collected during the working day is considered.Between two adjacent working days,the data fitting algorithm is used to fit the head and tail data.,Replace part of the data at the splicing area with fitting data,and then use the multi-layer long-term short-term memory neural network(LSTM)model to predict the processed data and the original data separately,compare the prediction results of the two,and finally prove that this paper proposes The method can get more accurate prediction results after processing the data.