End-to-end NILM model of industrial power data based on autoencoder transformer

Ce Li, Fanglin Guo, Rong Yang, Hua Wang , Bo Yao — Wed, 28 Feb 2024 00:00:00 +0000

Energy detection is an important part of intelligent power consumption, and its key technology is non-intrusive load monitoring (NILM). In this study, an end-to-end model is proposed to realize the NILM of commercial power data using the autoencoder-based transformer method. Firstly, we measured the operating power of different electrical appliances across different modes and combined the operating modes of electrical appliances. Considering the relatively large number of industrial electrical appliances, to ensure accuracy, we used Autoencoder to recode and reduce the dimension of the combined coding. Secondly, the transformer model was used to train the translation of the total power consumption information sequence and the state sequence of electrical appliances. Through our model, the electrical signals to be decomposed can be translated into different electrical state codes so as to realize load energy decomposition. Finally, when our model was applied to the gas station field data, the accuracy was as high as 90.17%.

Comparison of the elevator traffic flow prediction between the neural networks of CNN and LSTM

Mo Shi, Yeol Choi — Wed, 18 Dec 2024 08:33:33 +0000

With urbanization rapidly increasing, the demand for efficient elevator systems is becoming ever more pressing, particularly in crowded urban centers where high-rise buildings are prevalent. To solve this issue, elevator traffic analysis and prediction have emerged as critical components for optimizing elevator control systems. The elevator traffic flow prediction not only ensures smoother operations during peak usage times but also significantly reduces waiting periods for passengers, thereby enhancing overall convenience. By leveraging neural networks, the performance of elevator control systems is expected to be improved, leading to more efficient and convenient elevator utilization in both residential and commercial environments. Over the past few decades, the rapid advancements in neural networks have provided valuable tools for predicting traffic flows. In this research, a total of 655 actual ETF (Elevator Traffic Flow) data points from a typical office building on a weekday are utilized to analyze and predict traffic patterns using CNN (Convolutional Neural Networks) and LSTM (Long Short-Term Memory). The objective is not only to demonstrate the applicability of the neural networks in predicting elevator traffic flow but also to conduct a comparative analysis to identify which offers greater accuracy and suitability for the elevator traffic flow prediction. By enhancing the capabilities of elevator control systems through CNN or LSTM, this research seeks to improve not only the efficiency of elevator operations but also the overall living and working environment in urban cities. The findings from this research can inform subsequent research efforts, encouraging a deeper exploration of how synthetic predictions can further optimize elevator control systems, while the synthetic elevator control system is expected to lead to significant improvements in passenger experience, reducing wait times and increasing overall satisfaction in both residential and commercial buildings. Therefore, the insights gained from this research are expected to play a crucial role in shaping the future of smart buildings, aligning with the demands of modern urban living.

Intelligent Control and System Engineering

End-to-end NILM model of industrial power data based on autoencoder transformer

Comparison of the elevator traffic flow prediction between the neural networks of CNN and LSTM