As depicted in Figure X, the target value is divided into upper and lower halves. In the upper half, the predicted PF value approximates the target value well, while in the lower half, the predicted PF value is overly dependent on the observed value, resulting in a large error and unsatisfactory overall fitting effect. IPF reduces the prediction errors of both the upper and lower parts compared to PF. However, the predicted value of IPF is larger than the target value as a whole, even exceeding the maximum observed value. Nevertheless, it can preliminarily predict the change trend of the target value. The most significant difference between the predicted values of EPF and PF/IPF is that EPF seeks a balance between the observed value and the target value, relying heavily on the original observed value. Therefore, EPF greatly reduces the error of the observed value but cannot accurately predict the change trend of the target value. KF-PF is similar to EPF, but its error is reduced. In contrast, the proposed PSSB-PF algorithm demonstrates strong robustness in both accuracy and trend prediction. PSSB-PF has the best prediction effect in the part where the change trend of the target value is relatively flat. In the part where the curve trend is complex, PSSB-PF is affected by the observed value and appears as a peak value in the same direction as the observed value, but the overall effect is excellent. Compared with PF, IPF, EPF, and KF-PF, the prediction accuracy of PSSB-PF is improved by 29.67%, 55.33%, 61.33%, and 60.44%, respectively. Moreover, the accuracy of PSSB-PF is improved by 61.22% compared to the observed value.