Multi-unit soft sensing permits few-shot learning

Overview

• This paper explores ways to improve soft sensors using learning algorithms with transferability.
• It presents a probabilistic, hierarchical model implemented using a deep neural network for multi-unit soft sensing.
• The model is evaluated on a large-scale industrial case of developing virtual flow meters for 80 petroleum wells.
• The paper investigates how the model generalizes with the number of wells/units and demonstrates that multi-unit models learned from data from many wells enable few-shot learning of virtual flow meters for new wells.

Plain English Explanation

Soft sensors are devices that estimate or predict the value of a physical quantity, like flow rate, without directly measuring it. Instead, they use machine learning algorithms to make these estimates based on other easily measured quantities.

One way to improve soft sensors is to apply transfer learning - allowing the model to learn from related tasks and apply that knowledge to a new task. This is particularly relevant when developing soft sensors for similar, but physically different processes or units, like different petroleum wells.

In this paper, the researchers present a probabilistic, hierarchical model implemented using a deep neural network that can take advantage of this transferability for multi-unit soft sensing. They evaluate the model on 80 petroleum wells, and find that the multi-unit models learned from data across many wells can be used to quickly (in a "few-shot" manner) develop virtual flow meters for new wells. This is useful, as it can be difficult and time-consuming to collect enough data to train a soft sensor model for each new well.

Technical Explanation

The researchers formulate multi-unit soft sensing as a probabilistic, hierarchical model, which they implement using a deep neural network. This allows the model to take advantage of the transferability of knowledge between related soft sensor learning tasks, such as developing virtual flow meters for different petroleum wells.

The researchers evaluate their model on a large-scale industrial case, where they develop virtual flow meters (a type of soft sensor) for 80 petroleum wells. They investigate how the model's performance generalizes as the number of wells/units increases.

Interestingly, the researchers demonstrate that the multi-unit models learned from data across many wells enable few-shot learning of virtual flow meters for new wells. This means that the model can achieve high performance on new wells with just 1-3 data points, which is surprising given the difficulty of the task.

Critical Analysis

The paper provides a compelling approach for leveraging transferability to improve soft sensor development, particularly in the context of multi-task learning across similar units or processes.

One potential limitation is the reliance on a deep neural network, which can be computationally expensive and may require large amounts of data to train effectively. The researchers do not provide a detailed analysis of the model's complexity or training requirements.

Additionally, the paper focuses on a specific industrial case of virtual flow meters for petroleum wells. While the insights are likely applicable to other soft sensor development tasks, the researchers do not discuss the generalizability of their approach to different domains or sensor types.

Further research could explore the performance of the model on a wider range of soft sensor applications, as well as investigate more efficient or interpretable model architectures that could make the approach more accessible to practitioners.

Conclusion

This paper presents a novel approach for improving soft sensor development by leveraging transferability through a probabilistic, hierarchical model. The key finding is that multi-unit models learned from data across many wells can enable few-shot learning of virtual flow meters for new wells, significantly reducing the effort required to deploy soft sensors in new settings.

This research highlights the potential of transfer learning and multi-sensor fusion techniques to enhance the performance and efficiency of soft sensing applications, which have important industrial and scientific applications. Further advancements in this area could lead to more robust and adaptable soft sensors that can be more easily deployed in a wide range of real-world scenarios.