Efficient Multi-Channel Thermal Monitoring and Temperature Prediction Based on Improved Linear Regression

The early prediction and accurate tracking of the thermal characteristics of the CPU of a server can help avoid thermal failure and runaway due to defects in its thermal design. This study proposes a method of thermal monitoring and temperature prediction based on infrared thermocouples. A nine-channel whole-machine thermal monitoring system based on a thermal module is designed. It collects the real-time temperature data of each channel by using the Raspberry Pi platform through the (IC)-C-2 protocol and a visual digital interface that can intuitively display the dynamic distribution of the related temperatures. Based on data on the multi-channel historical thermal distribution as well as continuous training on real-time temperature data, an improved linear regression algorithm-based thermal prediction model is proposed to obtain rules governing the thermal characteristics of the CPU during operation. The results of experiments show that the accuracy of the temperature data collected using our proposed method of thermal monitoring can be controlled to within a range of error of 0.15%, with an average range of 0.62%. In addition, the improved linear regression algorithm predicts a good fitness with the empirical values, with a data coincidence of up to 0.96 compared with the traditional linear regression algorithm. The proposed method to monitor the thermal characteristics and predict the temperature of the CPU can provide a reference for the design and thermal diagnosis of high-performance servers.

Automated summary

This study proposes a thermal monitoring and temperature prediction method based on infrared thermocouples. A nine-channel whole-machine thermal monitoring system is designed to collect real-time temperature data. The improved linear regression algorithm can accurately predict the thermal characteristics of the CPU based on historical thermal distribution data and real-time temperature data. The experimental results show that the proposed method can accurately monitor temperature data and has a high coincidence with the actual values.