Method for predicting dynamic current-carrying capacity of transmission lines by integrating improved VMD and time-varying ensemble model

Accurately predicting the dynamic thermal rating of power lines is the core of overcoming the bottleneck in the engineering application of line dynamic capacity – increase technology. In response to the limitations of DTR data decomposition caused by the existing methods relying on manual experience to set the hyperparameters of the Variational Mode Decomposition algorithm, and the problems of single models, such as their inability to accurately extract multi – scale time – series features of DTR data and poor generalization ability, this paper proposes a transmission line DTR prediction method driven by an improved VMD and based on a time – varying multi – model ensemble. First, by constructing a fitness function based on the minimum envelope entropy of DTR data components, a mathematical mapping relationship between the hyperparameter space and the VMD decomposition effect is established. This transforms the traditional experience – oriented hyperparameter selection into a quantifiable optimization problem for solution. Subsequently, the global search ability of the slime mold algorithm is utilized to iteratively search for the optimal hyperparameters of VMD in the hyperparameter space. Second, by integrating the dynamic recursive characteristics of Elman and the multi – scale convolutional feature extraction advantages of TCN, a time – varying ensemble model is constructed to accurately extract the long – and short – term time – series features of DTR data. In particular, a dynamic weighting mechanism based on grey relational coefficients is designed. By quantifying the local correlation between the predicted values of the Elman model and the TCN model and the real values point – by – point, the weights can be adaptively allocated, effectively solving the problem of insufficient generalization ability of single models. Experimental results show that the proposed method maintains a prediction accuracy of over 95% in different datasets, which is 5.7% higher than that of the best traditional methods. It provides more theoretical support for the safe capacity increase of power lines and has significant engineering application value.