@article{scholars20140,
         journal = {IEEE Access},
       publisher = {Institute of Electrical and Electronics Engineers Inc.},
           pages = {35700--35715},
            year = {2024},
           title = {Optimized Multi-Level Multi-Type Ensemble (OMME) Forecasting Model for Univariate Time Series},
          volume = {12},
            note = {cited By 0},
             doi = {10.1109/ACCESS.2024.3370679},
          author = {Usmani, M. and Memon, Z. A. and Danyaro, K. U. and Qureshi, R.},
            issn = {21693536},
             url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85186967039&doi=10.1109\%2fACCESS.2024.3370679&partnerID=40&md5=34d476c35033ca2d0e3c33e62e853b54},
        keywords = {Economics; Electric load forecasting; Electric power distribution; Electric power utilization; Energy efficiency; Environmental impact; Learning systems; Long short-term memory; Tabu search; Time series, ARIMA; Deep learning; Energy; Energy-consumption; Ensemble learning; Ensemble methods; Exponential smoothing; GRU; Load modeling; LSTM; Machine-learning; MLP; Neural-networks; Optimisations; Predictive models; Smoothing methods; Time-series analysis; Times series; Univariate, Time series analysis},
        abstract = {Energy is of paramount importance for the world, and it is a fundamental driver of economic growth and development. Industries, businesses, and households rely on energy for even a small task. Due to its high demand, a significant portion of the global population still lacks access to reliable and affordable energy sources. Many industries and sectors continue to waste significant amounts of energy through inefficiencies. While energy is essential, the production and consumption of energy can have significant environmental consequences. Predicting power usage can help to significantly improve energy efficiency, reduce costs, enhance grid reliability, and minimize the environmental impact of energy consumption. In this work, a novel model named, the Optimized Multi-level Multi-type Ensemble (OMME) Forecasting Model is presented to estimate the power consumption. The proposed model was applied to the data set of total power consumption recorded in Austria at each hour after the pre-processing. The model applied bootstrapping at level 1 and a hybrid ensemble at level 2. Both ensemble methods utilized different algorithms for time series forecasting including ARIMA, Exponential smoothing, LSTM, GRU, and MLP. The parameters were tuned using Bayesian and Tabu search optimization. Different experiments were conducted for day and night usage separately. The proposed model was able to estimate the power usage with an error of 22. This work also learned the most suitable technique for power consumption time series. GRU performed very well in different experiments and gave the forecast with 12error. Distribution graphs of the OMME prediction further validate the integrity of the results. {\^A}{\copyright} 2013 IEEE.}
}