Single and Multi-Sequence Deep Learning Models for Short and Medium Term Electric Load Forecasting

Abstract

Time series analysis using long short term memory (LSTM) deep learning is a very attractive strategy to achieve accurate electric load forecasting. Although it outperforms most machine learning approaches, the LSTM forecasting model still reveals a lack of validity because it neglects several characteristics of the electric load exhibited by time series. In this work, we propose a load-forecasting model based on enhanced-LSTM that explicitly considers the periodicity characteristic of the electric load by using multiple sequences of inputs time lags. An autoregressive model is developed together with an autocorrelation function (ACF) to regress consumption and identify the most relevant time lags to feed the multi-sequence LSTM. Two variations of deep neural networks, LSTM and gated recurrent unit (GRU) are developed for both single and multi-sequence time-lagged features. These models are compared to each other and to a spectrum of data mining benchmark techniques including artificial neural networks (ANN), boosting, and bagging ensemble trees. France Metropolitan’s electricity consumption data is used to train and validate our models. The obtained results show that GRU- and LSTM-based deep learning model with multi-sequence time lags achieve higher performance than other alternatives including the single-sequence LSTM. It is demonstrated that the new models can capture critical characteristics of complex time series (i.e., periodicity) by encompassing past information from multiple timescale sequences. These models subsequently achieve predictions that are more accurate.

Multi-Sequence LSTM-RNN Deep Learning and Metaheuristics for Electric Load Forecasting

“…The performance of the optimized model is then compared to the benchmark. For our benchmark, we used random forest and support vector regressor machine learning models as well as a multi-sequence LSTM model from our previous work, whose inputs and hyperparameters have been established through experimentation [10]. The aforementioned multi-sequence model outperformed challenging machine learning models such as the ANN and the ensemble for this complex time series forecasting task.…”

“…Compared to shallow ANN architecture, deep learning models apply non-linear transformations to automatically learn complex temporal patterns via high-level abstractions. LSTM's high performance is achieved by memorizing long-term dependencies, thanks to an internal memory that makes it perfectly suitable for problems involving sequence-dependent behavior such as electricity load and demand [9,10]. In other terms, load forecasting characteristics make their predictions based on past observations since the pattern and the behavior of the energy consumption will likely reappear in the future.…”

“…In this paper, we customize the application of two evolutionary metaheuristics, namely a genetic algorithm (GA) and particle swarm optimization (PSO), to the problem of optimizing the performance of an LSTM model for electric load forecasting. Our approach is compared to alternative baseline approaches that use state-of-the-art machine learning techniques as well as a multi-sequence LSTM model with manually tuned parameters through extensive experimentation [9,10]. The comparison shows that the application of metaheuristics for the optimal configuration of electric load forecasting derived a multi-sequence LSTM model that performs significantly better than the benchmark models including other machine learning techniques (i.e., SVR, random forest, and ANN) and manually configurated LSTM.…”

Deep Learning with Stacked Denoising Auto-Encoder for Short-Term Electric Load Forecasting

“…The authors in [23] utilized the LSTM to establish a model considering a probabilistic load forecasting scenario in terms of quantiles. Salah Bouktif et al improved LSTM by using multiple sequences of relevant inputs time lags to incorporate periodicity characteristics of the electric load [24]. It turned out that the multi-sequence LSTM performs better than single-sequence LSTM.…”

A Weekend Load Forecasting Model Based on Semi-Parametric Regression Analysis Considering Weather and Load Interaction

“…For a long time, many experts have been focusing on research on short-term load forecasting. The forecasting methods such as the time series method [3,4], support vector machine method [5][6][7][8][9][10], random forest models [11][12][13][14], artificial neural network method [5,[15][16][17][18][19][20] and grey theory [21][22][23] could be applied to general weekday scenes and obtain good results. However, the difference between characteristics on weekend load and working days, as well as the interactive coupling relationship with external weather information, have become the shortcoming factors that restrict the accuracy of weekend load forecasting.…”