Join us on Zoom: https://ucsc.zoom.us/j/98444997043?pwd=TG9Yd0F6NGRET08ybXZiY2VCdUNTdz09 / Passcode: 629699
Description: The hidden information and characteristics embedded in electrical load profiles are indispensable for the effective planning and operation of power grids. Load forecasting plays a vitally important role in many applications for the electric industry, e.g., energy generation and transactions, load shedding and restoration, as well as infrastructure expansions. Based on historical data, accurate load forecasts provide a good reference for the needed load demand that can increase the efficiency and revenues for the electricity generation and distribution companies. In parallel with load forecasting, load monitoring can identify various types and statuses of loads by disaggregating the total power consumption into individual appliance levels. A scientific procedure of load monitoring facilitates the establishment of user-profiles, power usage habits, and peak load shifting. This is beneficial for both the end-users and utilities by improving the overall efficiency of the power network.
This research proposes a novel load forecasting paradigm and a dynamic data augmentation approach for non-intrusive load monitoring (NILM). First, by leveraging the state-of-the-art network structure of the bidirectional long short-term memory, we develop a load forecasting framework that consists of three interactive modules: i) a dynamic feature weighting module; ii) a load forecasting model with hierarchical temporal attention that incorporates similar day information; and iii) an error correction module that avoids extra designs and hyper-parameter tuning. The framework adopts a modular design with a good generalization capability, which is easy to customize and modify separately. Second, a dynamic data augmentation algorithm is proposed for the NILM to generate synthetic appliance profiles based on real data. Different from existing works that develop an individual model for each appliance, we utilize the multitask training technique to explore the shared knowledge and correlations among various appliances.
Extensive numerical simulations show that the proposed load forecasting framework outperforms a few existing forecasting methods. We demonstrate that the feature weighting mechanism and the error correction module are essential to achieve superior performance. Regarding the NILM task, the ablation study reflects that the proposed data augmentation and multitask training schema can further enhance the model generalization capability and performance.