Memetic Algorithms-based Electricity Load Forecasting Models With Applications

With the growing complexity of power system operating environment, and deepening reform of China’s competitive market-oriented electricity power system under China’s ’New Normal’ Economy, accurate load forecasting plays more and more important role in power systems operations, management, and commercial transactions. It has become an important research topic of management science in power system operations and management.The importance of load forecasting has motivated a wide variety of studies to develop forecasting methods, ranging from statistical ones such as autoregressive integrated moving average (ARIMA), exponential smoothing, and regression models, to artificial intelligence-based models such as neural networks (NNs) and support vector machines (SVMs). However, no matter which forecasting technique is applied, the performance are strongly influenced by many key technical issues, i.e., input feature selection, parameter selection, model selection, etc.. In this dissertation, some advanced learning algorithm such as support vector machines and multi-output support vector machines are applied for load forecasting. In order to build adaptive load forecasting model and to improve the forecasting performance, different memetic algorithms (MAs) are designed to solve the key technical problems. Experimental studies are conducted based on real-world applications for short-term load forecasting, mid-term load forecasting, and mid-term interval-valued load forecasting problems. In addition, Electricity Load Forecasting Support System (ELFSS) are designed based on the aforementioned studies.The main results of this dissertation are summarized as follows:Firstly, by focusing on the parameter selection problem of SVR-based load forecasting model, this study proposed a firefly algorithm (FA) based memetic algorithm (FA-MA) to appropriately determine the parameters of SVR forecasting model. Experimental results confirm that the proposed FA-MA can improve the forecasting performance of SVR-based load forecasting model by adaptively determining the adjustable parameters.Secondly, the selection of input features from a large pool of candidates is an important issue in the modeling of load forecasting.(1) Previous studies along this line of research have focused pre-dominantly on filter and wrapper methods. Given the potential value of a hybrid selection scheme that includes both filter and wrapper methods in constructing an appropriate pool of features, coupled with the general lack of success in employing filter or wrapper methods individually, in this study we propose a hybrid filter-wrapper approach for STLF feature selection. Two real-world electricity load datasets have been used to test the performance of the proposed approach, and the experimental results show its superiority over selected counterparts.(2) Instead of point load forecasting, this study proposed a multi-output support vector regression (MSVR) model for mid-term interval loads forecasting. In addition, a memetic algorithm (MA) based on the firefly algorithm is used to select proper input features among the feature candidates. Experimental results show that the proposed MSVR-MA forecasting framework may be a promising alternative for interval load forecasting.Thirdly, by evolving feature selection and parameter optimization simultaneously, a generalized model selection problem in load forecasting using SVR is proposed. To solve the proposed model selection problem, a comprehensive learning particle swarm optimization (CLPSO)-based memetic algorithm (CLPSO-MA) is designed. Compared with other well-established counterparts, benefits of the proposed model selection problem and the proposed CLPSO-MA for model selection are verified using two real-world electricity load datasets.Lastly, based on the above theoretical research achievement, an Electricity Load Forecasting Support System (ELFSS) is designed. The ELFSS provides the decision makers with the predicted future loads by using excellent load forecasting models and professional knowledge from domain experts. By integrating the research achievement we have made for parameter optimization, feature selection, and model selection in the above studies, the system has the capability of adaptively optimizing the load forecasting models and making predictions.