Temperature Prediction for Ultra-Low Temperature Measurement System Based on Improved Extreme Learning Machine Algorithm

Temperature monitoring is essential to ensure the operational safety of fusion devices in the ultra-low temperature liquid helium region. However, the unavoidable errors caused by environmental effects severely hinder the accuracy of its temperature measurement. Considering that traditional methods are not accurate enough to predict the temperature at low temperatures, for the first time, this work proposes an algorithm based on the SSA-PSO-ELM network for constructing a prediction model of the cryogenic measurement system. By this method, the initial number of hidden layer nodes is first randomly generated, then the network is trained with weight thresholds, and finally a quadratic optimization search is performed to determine the best network prediction model. Compared with the traditional algorithm, our algorithm effectively reduces the prediction error, improve the accuracy, and realize more accurate temperature monitoring in ultra-low temperature environment.