A combination model with variable weight optimization for short-term electrical load forecasting. (1st December 2018)

• Record Type: Journal Article
• Authors: Li, Wei-Qin; Chang, Li
• Is Part Of: Energy. Volume 164(2018)
• Page Start: 575
• Page End: 593
• Publication Date: 2018-12-01
• Subjects: Electrical power forecasting -- Combination model -- Culture particle swarm optimization -- Anomaly detection; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2018.09.027
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: The present study establishes a robust combination forecasting model and achieves the accurate prediction of electrical load by considering the dependency of the load series and the meteorological factors. On this basis, the culture particle swarm optimization algorithm is developed to improve the accuracy of the forecast. The merit is that by the particle mutation strategy, parameter adjustment strategy dependent on the fitness and the knowledge updating strategy, particles are avoided to trap in local optimum, consequently improving the computational speed and performance. Moreover, the data preprocessing technology based on the EEMD is proposed to reduce the random noises of the load series and to improve the robust of the forecasting model. The anomaly detection model is proposed in view of the probability distribution of relative errors. To assess the applicability and accuracy of the proposed model, it is compared with ant colony optimization, genetic algorithm, simulated annealing approach, cuckoo search algorithm, differential evaluation and artificial cooperative search. Results validated by the actual data sets for Shaanxi province, China, show higher accuracy and better reliability of the proposed model in comparison with other optimization models. Highlights: Develop a data preprocessing technique based on EEMD to reduce noise interference. Propose a CPSO method to optimize weights in combination models, promoting accuracy. Establish an anomaly detection model based on neural network, improving robustness.