A time series prediction model utilizing attribute correlation

Aiming at the problem that existing studies usually ignore the correlation between the attributes of time-series datasets, leading to poor model accuracy in long time-series prediction, a time-series prediction method utilizing attribute correlation is proposed. Firstly, the attribute grouping strategy of the time series dataset is designed to divide the attributes of the original dataset. Secondly, the features of the time series data in the split dataset are extracted by using the self-attention mechanism of ProbSparseCV with the fusion of coefficients of variation, and the prediction model is constructed by using the structure of the encoder-decoder. Finally, the validity and superiority of the proposed method are validated through several sets of experiments, the degree of contribution to the accuracy improvement of the proposed method is analyzed through the ablation experiments, and the improvement of long-time series prediction accuracy is realized.