A physics-enhanced multi-modal fused neural network for predicting contamination length interval in pipeline

Jian Du,Pengtao Niu,Jianqin Zheng,Qi Liao,Yongtu Liang

2024-09-19

Abstract:During the operation of a multi-product pipeline, an accurate and effective prediction of contamination length interval is the central key to guiding the cutting plan formulation and improving the economic effect. However, the existing methods focus on extracting implicit principles and insufficient feature correlations in a data-driven pattern but overlook the potential knowledge in the scientific theory of contamination development, may cause practically useless results. Consequently, in this study, the holistic feature correlations and physical knowledge are extracted and integrated into the neural network to propose a physics-enhanced adaptive multi-modal fused neural network (PE-AMFNN) for contamination length interval prediction. In PE-AMFNN, a multi-modal adaptive feature fusion module is created to establish a comprehensive feature space with quantified feature importance, thus capturing sufficient feature correlations. Subsequently, a mechanism-coupled customized neural network is designed to incorporate the explicit scientific principle into the forward and backward propagation. Besides, a physics-embedded loss function, which introduces interval differences and interval correlation constraints, is established to unearth the latent physical knowledge in contamination development and force the model to draw physically unreasonable results. Validation on the real-world cases implies that the proposed model outperforms the start-of-art techniques and latest achievements, with Root Mean Squared Relative Errors reduced by 31% and 36% in lower and upper limit prediction. Furthermore, the sensitivity analysis of model modules suggests that both the multi-modal feature fusion and the physical principle are crucial for model improvements

Computational Engineering, Finance, and Science

What problem does this paper attempt to address?

This paper attempts to solve the problem of how to accurately predict the pollution length interval during the operation of multi - product pipelines. Specifically, the paper focuses on the fact that when continuously transporting different types of refined oil products, the polluted products formed by the mixing interfaces between different oil products will lead to a decline in product quality, thereby causing economic losses and quality problems. In order to handle these polluted products, the transportation stations need to formulate cutting plans based on the prediction of the pollution length. However, the existing methods mainly rely on data - driven ways to extract implicit laws and feature correlations, while ignoring the scientific theoretical knowledge behind the pollution development, which may lead to inaccurate results in practical applications. Therefore, this paper proposes a physically - enhanced adaptive multi - modal fusion neural network (PE - AMFNN), aiming to improve the accuracy of pollution length interval prediction by integrating overall feature correlations and physical knowledge.

The main contributions of the paper are as follows:
1. Propose an adaptive multi - modal feature fusion strategy to capture the comprehensive correlation patterns between various input features and the pollution length range, and further quantify different influence mechanisms.
2. Design a custom neural network with mechanism coupling, integrate the transfer mechanism into the forward and backward propagation processes, and establish a coupled loss function, so that the neural network conforms to the explicit knowledge of the pollution length interval, thereby improving the interpretability and effectiveness of the model.
3. Propose the first physically - enhanced multi - modal fusion modeling paradigm for predicting the pollution length range from the perspective of practical professional knowledge, providing more interpretable and practical pollution development information.
4. Use the real - life pollution length range cases in multi - product pipelines to verify the effectiveness and accuracy of the proposed model, and illustrate the importance of each module through sensitivity analysis.

Through the above methods, the paper not only solves the deficiencies in the existing methods, but also provides a more accurate and reliable pollution length interval prediction tool for practical engineering applications.