Time series algorithm based multi mode switching method

Abstract: The rise in chip temperature will greatly reduce the lifespan of electronic devices, and the inability to detect faults in a timely manner will lead to a decrease in reliability. The commonly used solution is to cool down the chip as soon as possible during idle periods through sleep mode. In order not to affect device usage, these functions are scheduled to run when the system is idle. The operation of these functions is accompanied by some problems, such as the sudden start-up during sleep, which will cause significant fluctuations in system power and reduce the lifespan of the power module. Fault diagnosis during idle periods may affect device start-up. Therefore, how to make reasonable use of idle time, and how to arrange sleep and fault diagnosis modes during system idle periods have become important research topics. This article conducts research on pattern management for data acquisition systems that work repeatedly. The research modes include normal working mode, shallow sleep mode, deep sleep mode, and self check mode. Different mode management schemes have been designed for different lengths of idle time. In addition, an indicator was designed to evaluate the effectiveness of pattern management, which is the effective management rate. The effective management rate is the probability that the system correctly selects the pattern management scheme and successfully executes the corresponding pattern. Based on the idle time prediction algorithm, pattern management is implemented. The appropriate pattern management scheme is selected according to the predicted idle time length. The pattern management strategy is tested and improved according to the experimental results. By combining different thresholds and weights for testing, the optimal combination of thresholds and weights is found, resulting in an effective management rate of 90.34%.