Research on Short-Term Load Prediction Based on Seq2seq Model

Abstract

Electricity load prediction is the primary basis on which power-related departments to make logical and effective generation plans and scientific scheduling plans for the most effective power utilization. The perpetual evolution of deep learning has recommended advanced and innovative concepts for short-term load prediction. Taking into consideration the time and nonlinear characteristics of power system load data and further considering the impact of historical and future information on the current state, this paper proposes a Seq2seq short-term load prediction model based on a long short-term memory network (LSTM). Firstly, the periodic fluctuation characteristics of users’ load data are analyzed, establishing a correlation of the load data so as to determine the model’s order in the time series. Secondly, the specifications of the Seq2seq model are given preference and a coalescence of the Residual mechanism (Residual) and the two Attention mechanisms (Attention) is developed. Then, comparing the predictive performance of the model under different types of Attention mechanism, this paper finally adopts the Seq2seq short-term load prediction model of Residual LSTM and the Bahdanau Attention mechanism. Eventually, the prediction model obtains better results when merging the actual power system load data of a certain place. In order to validate the developed model, the Seq2seq was compared with recurrent neural network (RNN), LSTM, and gated recurrent unit (GRU) algorithms. Last but not least, the performance indices were calculated. when training and testing the model with power system load data, it was noted that the root mean square error (RMSE) of Seq2seq was decreased by 6.61%, 16.95%, and 7.80% compared with RNN, LSTM, and GRU, respectively. In addition, a supplementary case study was carried out using data for a small power system considering different weather conditions and user behaviors in order to confirm the applicability and stability of the proposed model. The Seq2seq model for short-term load prediction can be reported to demonstrate superiority in all areas, exhibiting better prediction and stable performance.

Additive Ensemble Neural Network with Constrained Weighted Quantile Loss for Probabilistic Electric-Load Forecasting

“…Seq2seq models: The sequence-to-sequence models were initially used in NLP, but their applicability has spread to almost any time-series forecasting problem. Until now, these models have been little applied to STLF, and their results have been good for very short-term forecasts [ 52 ].…”

“…The ML and DMD models expect a sequence of scalar values (longitudinal data) as input; the way to transform the input data for these models is to flatten the vectors over all time-steps. The DL models can receive vector-valued inputs, i.e., both LSTM [ 52 ] and 1D/2D CNN [ 51 ] models can receive a vector-valued sequence (with length p) where each timestep is represented by a vector of values. When the first layer of the DL model is an FC layer, the input must be formatted as a vector (flattened), and when the first layer is a 2D-convolutional (2D-conv) layer, the data must be formatted as a matrix by packing the predictors (in columns) for all past time-steps (in rows).…”

“…The conclusions obtained in this work are consistent with the results obtained by the present study. A similar study is presented in [ 52 ], where research is conducted comparing a Seq2seq model (with and without attention) with alternative DL models based exclusively on different types of recurrent networks, such as long short-term memory networks (LSTM) and gated recurrent unit (GRU). In this case, the Seq2seq model presents the best results for short-term forecasting, also following the results obtained in the present work.…”

Analysing and Forecasting Electricity Demand and Price Using Deep Learning Model During the COVID-19 Pandemic

“…A sequence-tosequence RNN was developed with an attention mechanism for the electric load forecast in recent research [15], and a similar sample generation method was designed. In [16], A LSTM-based short-term load forecast model with two mechanisms is built. Their approach is similar to the language translation model in [2], where one LSTM is used to encode the input sequence into a fixed vector, and then need separate LSTM to decode the vector to a sequence of outputs with an attention mechanism to learn weight.…”

Ensemble Residual Networks for Short-Term Load Forecasting

“…The method successfully captures the dependencies between these sequences. A Seq2seq short-term load forecast model based on LSTM is developed in [16]. Two deep learning methods were proposed for electric load forecasting in [17].…”