A Two-Stage Multistep-Ahead Electricity Load ForecastingScheme Based on LightGBM and Attention-BiLSTM

Optimized EWT-Seq2Seq-LSTM with Attention Mechanism to Insulators Fault Prediction

Insulators installed outdoors are vulnerable to the accumulation of contaminants on their surface, which raise their conductivity and increase leakage current until a flashover occurs. To improve the reliability of the electrical power system, it is possible to evaluate the development of the fault in relation to the increase in leakage current and thus predict whether a shutdown may occur. This paper proposes the use of empirical wavelet transform (EWT) to reduce the influence of non-representative variations and combines the attention mechanism with a long short-term memory (LSTM) recurrent network for prediction. The Optuna framework has been applied for hyperparameter optimization, resulting in a method called optimized EWT-Seq2Seq-LSTM with attention. The proposed model had a 10.17% lower mean square error (MSE) than the standard LSTM and a 5.36% lower MSE than the model without optimization, showing that the attention mechanism and hyperparameter optimization is a promising strategy.

Optimized LightGBM Power Fingerprint Identification Based on Entropy Features

The huge amount of power fingerprint data often has the problem of unbalanced categories and is difficult to upload by the limited data transmission rate for IoT communications. An optimized LightGBM power fingerprint extraction and identification method based on entropy features is proposed. First, the voltage and current signals were extracted on the basis of the time-domain features and V-I trajectory features, and a 56-dimensional original feature set containing six entropy features was constructed. Then, the Boruta algorithm with a light gradient boosting machine (LightGBM) as the base learner was used for feature selection of the original feature set, and a 23-dimensional optimal feature subset containing five entropy features was determined. Finally, the Optuna algorithm was used to optimize the hyperparameters of the LightGBM classifier. The classification performance of the power fingerprint identification model on imbalanced datasets was further improved by improving the loss function of the LightGBM model. The experimental results prove that the method can effectively reduce the computational complexity of feature extraction and reduce the amount of power fingerprint data transmission. It meets the recognition accuracy and efficiency requirements of a massive power fingerprint identification system.

Multi-Step Hourly Power Consumption Forecasting in a Healthcare Building with Recurrent Neural Networks and Empirical Mode Decomposition

Short-term forecasting of electric energy consumption has become a critical issue for companies selling and buying electricity because of the fluctuating and rising trend of its price. Forecasting tools based on Artificial Intelligence have proved to provide accurate and reliable prediction, especially Neural Networks, which have been widely used and have become one of the preferred ones. In this work, two of them, Long Short-Term Memories and Gated Recurrent Units, have been used along with a preprocessing algorithm, the Empirical Mode Decomposition, to make up a hybrid model to predict the following 24 hourly consumptions (a whole day ahead) of a hospital. Two different datasets have been used to forecast them: a univariate one in which only consumptions are used and a multivariate one in which other three variables (reactive consumption, temperature, and humidity) have been also used. The results achieved show that the best performances were obtained with the multivariate dataset. In this scenario, the hybrid models (neural network with preprocessing) clearly outperformed the simple ones (only the neural network). Both neural models provided similar performances in all cases. The best results (Mean Absolute Percentage Error: 3.51% and Root Mean Square Error: 55.06) were obtained with the Long Short-Term Memory with preprocessing with the multivariate dataset.

Interpretable Short-Term Electrical Load Forecasting Scheme Using Cubist

Daily peak load forecasting (DPLF) and total daily load forecasting (TDLF) are essential for optimal power system operation from one day to one week later. This study develops a Cubist-based incremental learning model to perform accurate and interpretable DPLF and TDLF. To this end, we employ time-series cross-validation to effectively reflect recent electrical load trends and patterns when constructing the model. We also analyze variable importance to identify the most crucial factors in the Cubist model. In the experiments, we used two publicly available building datasets and three educational building cluster datasets. The results showed that the proposed model yielded averages of 7.77 and 10.06 in mean absolute percentage error and coefficient of variation of the root mean square error, respectively. We also confirmed that temperature and holiday information are significant external factors, and electrical loads one day and one week ago are significant internal factors.