Short-term load forecasting is an important task for the planning and reliable operation of power grids. High-accuracy forecasting for individual customers helps to make arrangements for generation and reduce electricity costs. Artificial intelligent methods have been applied to short-term load forecasting in past research, but most did not consider electricity use characteristics, efficiency, and more influential factors. In this paper, a method for short-term load forecasting with multi-source data using gated recurrent unit neural networks is proposed. The load data of customers are preprocessed by clustering to reduce the interference of electricity use characteristics. The environmental factors including date, weather and temperature are quantified to extend the input of the whole network so that multi-source information is considered. Gated recurrent unit neural networks are used for extracting temporal features with simpler architecture and less convergence time in the hidden layers. The detailed results of the real-world experiments are shown by the forecasting curve and mean absolute percentage error to prove the availability and superiority of the proposed method compared to the current forecasting methods.

Intelligent power management system is an important part of the smart grid, in which the non-intrusive load monitoring technology is one of the key technologies. However, most of the load monitoring methods regard this task as a multi-classification problem, thus it is not effective to identify the unknown loads that did not participate in training. In this paper, an adaptive non-intrusive load monitoring method based on feature fusion has been proposed which utilizes both information of harmonic current feature and voltage-current (V-I) trajectory feature. The harmonic current feature is obtained from the high-frequency load sampling data through Fast Fourier Transform (FFT). At the same time, the V-I trajectory feature is obtained by using the pre-trained convolutional autoencoder neural network that was trained for feature extraction on a public dataset. After the feature extraction, the similarity between these two feature vectors and the available feature vectors in the database is calculated by TOPSIS algorithm. Then the load monitoring can be carried out according to the similarity. When the maximum similarity is greater or equal than the set threshold, the load is considered to be one of the existing loads in the database, otherwise, it will be considered as a new type of load, and the load feature database will be updated. The autoencoder model is trained by using the V-I trajectory from the BLUED dataset and the PLAID dataset is used to verify the identification accuracy of the proposed algorithm by comparison with different algorithms. On the PLAID dataset, the identification accuracy can reach above 97%. Finally, on the embedded Linux system with STM32MP1 as the core, some household electrical appliances are used for validation in real house environment. The results show that the proposed method has improved the capability of load identification by using the complementarity of different features. It can be carried out in real time on embedded system and is able to identify the unknown load at the same time.

The traditional non-intrusive load monitoring (NILM) algorithms are mostly based on classification models, which have several deficiencies. Firstly, a large amount of labeled data is required to train the classification model. Secondly, these algorithms cannot identify unknown devices that frequently encountered in practical application. Finally, these models have poor performance in versatility, which means they only adapt to the trained data. These shortcomings greatly influence the practicality of these NILM algorithms. To tackle these problems, this paper has proposed a non-intrusive adaptive load identification model based on the Siamese network, which uses both the V-I trajectory and active power as the load signatures. The Siamese network is utilized to calculate the similarity of the V-I trajectory, and the load identification is realized by matching the signature with the feature library. Through adding new features to the feature library dynamically, the identification of unknown load can be realized. In addition, the Siamese network is a typical network for few-shot learning, thus the proposed model can be trained with a small number of samples to achieve ideal recognition effect. At last, the validity and versatility of the model are verified in PLAID dataset and COOLL dataset.

In this paper, an optimal multi-timescale demand side scheduling framework, i.e., the combination of week-ahead and day-ahead, for industrial customers is proposed. Different demand side management (DSM) techniques suitable for distinct week-ahead and day-ahead timescales cooperate for achieving the overall optimal demand scheduling in the entire multi-timescale frame. Specifically, in the week-ahead scheduling, a dynamic scenario generation method is proposed to accurately simulate uncertainties of customer electricity demand time-series during the scheduling horizon, which can represent not only the marginal distribution of possible customer loads at each time instant but also the joint distribution among multiple loads at different time instants. In addition, priorities of various DSM techniques accepted by DSM participants and their willingness are also considered, aiming at mitigating impacts on their normal manufacturing process. With actual historical load data of industrial customers from advanced metering infrastructure system, the dynamic scenario generation method is shown to be effective in preserving statistic features of load fluctuations, and the proposed optimal multi-timescale coordinated demand side scheduling model is demonstrated to be an effective DSM approach.

Hydrogen energy storage system (HESS) has attracted tremendous interest due to its low emissions and high storage efficiency. In this article, the HESS is considered as an essential tool in hydrogen-integrated transportation and power systems to alleviate EV charging demand forecast error in a fast-charging station (FCS) and to solve voltage deviation problem due to the huge uptake of fast chargers on the utility grid. First, the wavelet transform (WT) method and long short-term memory (LSTM) neural network are combined to precisely predict the nonstationary traffic flow (TF). Then, a queueing theory-based model is developed to convert the predicted TF to the expected EV charging demand in FCS by considering charging service limitations and driver behaviors. Third, the charging demand prediction error is used to schedule the components in a HESS by considering their inherent properties and operational limits. As a result, the HESS configuration can be determined by analyzing the tradeoff between the investment cost and the monetary penalty due to charging demand forecast error and voltage deviation. The proposed solution is validated through a case study with mathematical justifications and simulation results.