基于神经网络的指标预警研究

Research on index early warning based on neural network

在平台使用和实际运维过程中,经常发生指标已经劣化,平台使用人员和运维人员才知道的情况。本文基于神经网络模型,以PM数据为基础,对小区历史数据进行研究。通过LSTM探寻小区在历史时间段指标的变化规律,以实现对未来指标的合理预测,从而提前预警问题指标。研究中进行了数据清洗和深度学习框架选择,并建立了LSTM模型进行指标预测。模型训练后,可将平均干扰电平值的误差控制在0.3 dBm内,并能对小区平均干扰电平值进行预测。该研究具有广泛的应用前景,适用于PM数据中数值型字段所指的指标,且可使用不同时间粒度的PM数据进行预测。

Based on the neural network model and PM data,this paper conducts research on the historical data of the community.Through the LSTM neural network,it explores the change rules of the indicators in the historical period of the community to achieve a reasonable prediction of future indicators and thus provide early warning of problematic indicators.In the study,data cleaning,selection of deep learning frameworks were carried out,and an LSTM neural network model was established for indicator prediction.After model training,the error of the average interference level value can be controlled within 0.3 dBm,and the average interference level value of the community can be predicted.This research has broad application prospects,is applicable to the indicators indicated by the numerical fi elds in PM data,and can use PM data of diff erent time granularities for prediction.