Abstract
From the perspective of energy providers, accurate short-term load forecasting plays a significant role in the energy generation plan, efficient energy distribution process and electricity price strategy optimisation. However, it is hard to achieve a satisfactory result because the historical data is irregular, non-smooth, non-linear and noisy. To handle these challenges, in this work, we introduce a novel model based on the Transformer network to provide an accurate day-ahead load forecasting service. Our model contains a similar day selection approach involving the LightGBM and k-means algorithms. Compared to the traditional RNN-based model, our proposed model can avoid falling into the local minimum and outperforming the global search. To evaluate the performance of our proposed model, we set up a series of simulation experiments based on the energy consumption data in Australia. The performance of our model has an average MAPE (mean absolute percentage error) of 1.13, where RNN is 4.18, and LSTM is 1.93.
Highlights
Anbarjafari (Shahab)With the advancement of the smart grid and the growing demand for economically efficient electricity scheduling, short term load forecasting (STLF) has attracted more attention in both industry and academia
We proposed a novel similar day selection approach that combines LightGBM and K-means algorithm to calculate the importance of each feature and select the similar days from historical time series data
root mean square error (RMSE) is a typical indicator of the regression model, which indicates how much error the model will produce in the prediction
Summary
With the advancement of the smart grid and the growing demand for economically efficient electricity scheduling, short term load forecasting (STLF) has attracted more attention in both industry and academia. Conventional statistical methods often involve double seasonal Holt-Winters exponential smoothing [4], linear regression, auto-regressive integrated moving average [5] etc These approaches do not always work satisfactorily as these models adopt linear functions to process the relationship between the actual and forecasted data. Compared with traditional statistical approaches, the artificial neural network method is a data-driven approach that has outstanding learning ability and adaptive function. This characteristic is very suitable for the time series forecasting problem. Compared to the standard recurrent neural network, there are two transfer states in LSTM It only remembers the essential information, which leads to outstanding performance over long sequences of input.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
