| Summary: | 碩士 === 國立中興大學 === 資訊管理學系所 === 100 === ABSTRACT
Nowadays, it is essential to consume electricity more efficiently and to understand customer usage habits. Since the precise load forecasting plays an important role in reducing unnecessary consumption and in arranging the dispatch of electric appliance. In other words, by installing a smart meter and electricity management systems, we can simulate electricity load forecast of the next day and adjust unnecessary electric appliances load to achieve reasonable energy-saving management for minimizing electricity expenditure of a single-person family. We can apply the forecasted load to better manage the use of electricity. Thus, carbon reduction of the society and energy efficiency will be realized. With Time-of-Use (TOU) framework, which records customer’s electricity usage of time, precise STLF provides suggestions to help people to save on electricity bills.
This paper is concerned with short-term load forecast (STLF) in a single-person family. It presents an integrated fuzzy system, data mining and neural network framework to estimate and predict electricity demand for real-time and daily changes in electricity consumptions. Also, the impact of data preprocessing and postprocessing methods on the fuzzy system performance is considered in this paper. A new fuzzy neural framework enhances the performance of the system as well as making the forecast more accurate considering the realistic conditions. According to the features of short-term power load, the influence of weather conditions and life styles are also considered in this study to extract the relationship between the electricity consumptions.
Lastly, the method is simulated using MATLAB, taking the load data from a single-person family in Tainan’s Yongkang City and the weather data from the Central Weather Bureau as an example. There are five cases that are comparison of different parameters, comparison of various days, comparison of various methods, comparison of various preprocessing methods and membership function optimized. After membership function optimizing, the results showed that the proposed new hybrid fuzzy neural system provides good predictive ability at MAPE 3.68%.
|
|---|
