Multi-step carbon price forecasting based on a new quadratic decomposition ensemble learning approach

Numerous studies show that it is reasonable and effective to apply decomposition technology deal with the complex carbon price series. However, existing research ignores residual term containing information after applying single technique. Considering demand for higher accuracy of series prediction following path, this paper proposes a new hybrid model VMD-CEEMDAN-LSSVM-LSTM, which combines quadratic technique optimized long short memory (LSTM). In part model, original processed by variational mode (VMD), then obtained further decomposed complete ensemble empirical adaptive noise (CEEMDAN). least squares support vector machine (LSSVM) introduced, LSSVM-LSTM constructed predict components decomposition. The selects two different case data from European Union emissions trading system (EU ETS) as samples. Taking results Case Ⅰ in 1-step ahead forecasting scenario an example, evaluation indexes eMAPE , id="m2">eRMSE id="m3">R2 VMD-CEEMDAN-LSSVM-LSTM are 0.3087, 0.0921 0.9987 respectively, significantly better than other benchmark models. confirm superiority robustness proposed forecasting.