KEYWORDS: Neural networks, Data modeling, Artificial neural networks, Power grids, Power consumption, Modal decomposition, Mathematical optimization, Machine learning, Education and training
In recent years, countries around the world have been advocating green development and gradually implementing new energy-based power grids. Due to the contradiction between the randomness of new energy generation and the growing demand for residential electricity, the stable operation of power grids is facing a huge challenge. Air conditioning load accounts for a large proportion of residential electricity consumption, so it is of great significance to monitor it. Through understanding the electricity consumption of air conditioning load, residents can be guided to use electricity reasonably, ensure the stable operation of the power grid, and provide a decision-making basis for demand response to alleviate the contradiction between supply and demand. At present, there are relatively few air conditioning load monitoring works, most of which are carried out as part of non-intrusive load monitoring. The selected input data mainly consider generality and seldom consider the operation rule of air conditioning and the influence of meteorological factors. Therefore, the accuracy of air conditioning monitoring is still insufficient. Based on this, a real-time monitoring method of air conditioning based on electrical characteristics and a twin-tower neural network is proposed in this paper. Based on Long Short-Term Memory (LSTM), the sequential branch was constructed to mine the sequential characteristics of air conditioning operation. Based on Back Propagation (BP), an electrical feature branch was constructed to capture the operation rules of air conditioning. Meanwhile, the experimental results on one year of real data from 20 users show that compared with the traditional LSTM model, this model can realize the accurate identification of air conditioning load and provide support for the demand response potential assessment of air conditioning load.
In recent years, the use of rooftop photovoltaic (PV) has increased as countries upgrade their energy systems. However, estimating the impact of behind-the-meter PV on grid operation requires precise physical models and weather information, which is not practical. To address this issue, we propose a data-driven approach using a sequence-to-subsequence (Seq2subseq) PV decomposition model based on the one-dimensional convolutional neural network (1D-CNN). This model automatically extracts temporal features from net metered sequences and outputs behind-the-meter PV generation using a sliding window. We evaluated our model on 184 rooftop PV users in the SGSC dataset, demonstrating its accuracy and ability to generalize across different climates. Our proposed approach offers an effective solution for real-world applications.
In recent years, load identification technology has received great attention as the value of real-time load-side electricity information has gradually emerged. There are several ways to precisely identify the different types of loads. However, practical situations with novel load types and little labeled data are seldom considered. For this reason, this paper proposes a few-shot identification method for novel loads based on the Model-Agnostic Meta-Learning (MAML). It uses the Adaptive Weighted Recurrence Graphs (AWRG) model as the base learner, which has the best performance in load identification, and pre-trains the model with existing data. The proposed method uses meta-training to get initial parameters that are generalized across multiple load types to improve the learning ability of the model on few-shot tasks with novel loads. Compared with transfer learning methods commonly used for generalized load identification, the results on the WHITED dataset show that the proposed method can improve the scalability of the load identification for practical applications.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.