ALTM: Adaptive learning-based thermal model for temperature predictions in data centers

Research advances on AI-powered thermal management for data centers

“…Some researchers have proposed their own thermal models. For instance, Mirhoseininejad et al [29] proposed a holistic thermal model, namely, the adoptive learningbased model (ALTM), which can predict the temperature of critical thermal zones using DC operational variables as inputs and outputs. Unlike many proposed thermal models that cannot dynamically change with changes within a data center, this model can be dynamically changeable as the nature of data centers.…”

“…Moreover, the Gaussian mixture model and log-linear model can solve the optimization problem. Mirhoseininejad et al [29] proposed an ALTM that quickly adapts to thermal changes in the data center environment without any prior awareness of heat transfer rules between data center entities, besides its ability to be efficiently used for either cooling system controllers or thermal-aware workload schedulers. For training the ATLM, the authors utilized a MATLAB toolkit in which the standard backpropagation method employs the Levenberg-Marquardt algorithm to train the model [29] .…”

“…Their results confirm the accuracy and adaptability of the model. The authors also explored weighted recursive least squares for the estimation of a linear model [29] .…”