TASE-Net: A Short-Term Load Forecasting Model Based on Temperature Accumulation Sequence Effect

Electricity consumption forecasting plays an important role in ensuring efficient dispatch and reliability of the grid. The results are influenced by several factors at the same time. Inspired by the effect of temperature accumulation on load which the load forecast is effected by the temperature of the previous days in a specific temperature range, in this paper, we propose a model structure based on temperature accumulation sequence effects. It incorporates the temperature accumulation effects in a network: Temperature Accumulation Sequence Effects Network(TASE-net) which in a way generates a set of temperature accumulation sequences, uses a combined K-Shape-PSF method for feature extraction, and abstracts the sequence identity by Temporal Convolutional Network (TCN). To verify our proposed method, it is compared with other state-of-the-art methods for extracting similar sequences by using the datasets from three regions. The experimental results show that TASE-net reduces the error by 16% to the comparative method and achieve better MAPE.