Short Term Power Load Forecasting Method Considering Data Quality Problems

Accurate short-term power load forecasting is conducive to maintaining the balance of power and electricity and the formulation of power grid dispatching plan, and is critical to the safe and stable operation of power system. However, due to external factors such as communication or hardware equipment failure, the power load data has certain data quality problems, which seriously affects the accurate prediction of power load. Aiming at this problem, a method of power load forecasting considering data quality is proposed. Firstly, an enhanced denoising autoencoder based on bidirectional long short-term memory is proposed to extract the time-series features that are robust to bad data. Then, an end-to-end mixture model is established by using the robust time series as the input of the temporal convolutional neural network to realize the accurate load forecasting. Finally, using the actual power load data, it is verified that the proposed method still has high prediction accuracy and applicability in the scenario with data quality problems.