Session: 07-10: Simulations and Predictions - II

Paper Number: 134843

134843 - Prediction of Sensor Data Accuracy in Thermal Experimental Benches Using Gru-Gcn Neural Network Model 

Abstract: 

In this research, we endeavor to significantly enhance the accuracy of sensor data prediction in thermal experimental bench by introducing a groundbreaking neural network model, known as the GRU-GCN. The early phase of our study concentrates on providing a comprehensive overview of the experimental bench's structure and its pivotal sensors. This foundational step is crucial for thoroughly understanding the inherent characteristics of the data, which in turn, informs the subsequent design and implementation of our model. Following this, the research pivots to a deep exploration of the GRU-GCN model itself. We meticulously examine its architectural framework, delve into the nuances of its training methodologies, and scrutinize the strategies employed for data processing. The model is specifically tailored for the analysis of spatiotemporal data, equipped with advanced capabilities to effectively capture and intricately analyze the time-series characteristics inherent in sensor data, as well as their complex interrelationships within the network structures. The uniqueness of the GRU-GCN model lies in its innovative integration of the Gated Recurrent Unit (GRU) and Graph Convolutional Network (GCN). The GRU component provides the model with the ability to efficiently process time-dependent data, allowing for a more nuanced understanding of temporal dynamics. Meanwhile, the GCN aspect facilitates the understanding of spatial relationships and patterns within the data, making the model exceptionally adept at handling the spatial intricacies of sensor networks. To validate the effectiveness of our model, we initiated a series of experiments on the thermal experimental bench. These experiments were designed to test the model's performance in real-world scenarios, simulating various conditions that sensors on the bench might encounter. The data collected from these experiments served as the primary dataset for training and testing our model. In an effort to underscore the superior performance of the GRU-GCN model, we engaged in a thorough comparative analysis with traditional single neural network models. This rigorous comparative study not only highlights the GRU-GCN's exceptional prowess in managing high-dimensional and intricate data sets but also illuminates its considerable strides in elevating prediction accuracy. Furthermore, we conducted extensive performance evaluations of the model under a diverse array of parameters and configurations. This aspect of our research was instrumental in gaining an enriched understanding of the model's versatility, adaptability, and its overall robustness in various operational scenarios. We systematically altered parameters such as learning rate, number of layers, and batch size to observe the model's response and performance under different conditions. Additionally, we integrated feedback mechanisms into the GRU-GCN model, enabling it to self-adjust and optimize its performance over time. This aspect of the model makes it not only powerful in its current form but also ensures its continual improvement and relevance in the face of evolving data patterns and environmental conditions. Concluding the study, we encapsulate the promising potential of the GRU-GCN model in the realm of predictive applications for sensor data on thermal experimental bench. We also chart out prospective avenues for future research. These envisioned pathways include initiatives to augment the model's generalization capabilities and to amplify its efficacy in processing increasingly complex and variable data sets.

Presenting Author: Linjun Yang Harbin Engineering University

Presenting Author Biography: Yang Linjun is engaged in research on artificial intelligence algorithms in the field of nuclear engineering.

Authors:

Linjun Yang Harbin Engineering University
Tong Li Harbin Engineering University
Yongchao Liu Harbin Engineering University
Bo Wang Fundamental Science on Nuclear Safety and Simulation Technology Laboratory, Harbin Engineering University, Harbin 150001, China
Jiangkuan Li Harbin Engineering University
Jiming Wen Harbin Engineering University
Sichao Tan Harbin Engineering University
Ruifeng Tian Harbin Engineering University, Harbin 150001, China