Abstract: The residual stress in high-temperature alloy ring rolled parts can affect the subsequent processing performance and service life of the components, quantitative detection of which in high-temperature alloy ring-rolled components is a challenging task in non-destructive testing applications. To improve the accuracy of Barkhausen noise quantitative detection of residual stress, tensile specimens were fabricated using high-temperature alloy GH2907 as the parent material. Barkhausen noise signals were collected in situ under loading conditions, and 15 Barkhausen signal characteristics such as root mean square and pulse factor were extracted. Kernel principal component analysis was employed to reduce the dimensionality of the detection signal feature data. The result of dimension-decrease and the corresponding average stress of the specimens were used to construct a sample set. A continuous variable regression model between Barkhausen signal features and residual stresses was established by combining particle swarm optimization (PSO) with support vector regression (SVR) to optimize parameters. With actual workpieces as detection objects, and the residual stress results measured by blind hole method as comparison, the effectiveness and accuracy of the predictive model were validated. The results showed that the goodness of fit of the established stress prediction model is 0.9318, and the average relative error of residual stress evaluation for the test samples was 7.1%. Compared with the linear evaluation model of Barkhausen residual stress with a single parameter, the PSO-SVR stress prediction model had higher goodness of fit, smaller error, and better generalization performance, enabling the detection and high-precision evaluation of residual stresses.