Temperature control and monitoring is a critical part of daily operations at large facilities, and is currently an urgent topic at NSLS-II. NSLS-II storage ring magnet cooling-water systems have developed blockages that reduce cooling, increase temperatures and endanger magnets, and so monitoring and preventative maintenance is critical. Machine learning algorithms provide a promising approach to giving Operators and system experts more intelligent feedback about temperature trends, to better guide preventative maintenance. In this paper, we construct two types of linear regression methods to predict temperature rise in QM Quadrupole magnets during daily operation: (1) piecewise linear regression and (2) overlapped piecewise linear regression. In (1), temperature data is collected, and piecewise regression is performed daily on 24-hours of data, to obtain a corresponding linear-fi?t slope. The slope parameter will be collected daily to generate a new rate-of-change parameter, using an accumulated regression model. In (2), the same regression techniques are used, except 24-hour dataset regression is performed every 12 hours (meaning 12 hrs of data overlap in neighboring datasets). Regression slopes are now found at twice the rate, giving a more accurate rate-of-change prediction. Di?fferent ratios of training part to predicting part are also studied to verify the quality of the methods.