NEC Corporation (hereinafter referred to as NEC), the National Institute of Advanced Industrial Science and Technology (hereinafter referred to as AIST), Mitsui Chemicals, Inc. (hereinafter referred to as Mitsui Chemicals), and Omega Simulation Co., Ltd. (hereinafter referred to as Omega Simulation) have built a plant operation support system that combines "AI for plant operation support" with a mirror plant reproduced on a simulator, successfully assisting with the startup operation at the large boiler plant of Mitsui Chemicals' Osaka factory.

As a result, operational support for transient states that are difficult to automate, such as the startup operations of boiler plants and chemical plants or changes in production volume, as well as reducing the time taken to reach steady-state operation from cold and warm shutdowns, leading to reductions in raw materials and Energy, and overall operational efficiency is anticipated.

Boiler plants in chemical factories are large and critical infrastructures that supply the necessary Electricity and steam for product production to all plants within the factory. Therefore, prompt resumption of operations is required as soon as the factory's regular maintenance is completed. In particular, if the resumption of operations is delayed, it can affect the startup of the plants that were halted concurrently with the regular maintenance of the boiler plant, thus influencing the production of each product.

Compared to controlling plants in steady operation, operating operations during startup, where temperature and pressure change significantly, are challenging and developing control systems has also been difficult. If heating is done rapidly in an attempt to quickly start a halted boiler plant, it can lead to damage to plant equipment or compromise the safety of operators. Therefore, while raising the temperature and pressure of the boiler, it is essential to operate quickly and efficiently without exceeding the prescribed speed, requiring advanced technology. Moreover, such large-scale plants are often operated 24 hours a day, excluding regular maintenance, and startup opportunities are limited, posing challenges in training experienced operators.

NEC and AIST have developed an AI technology called "Plant Operation Support AI", which can efficiently operate large-scale infrastructures such as plants and provide the rationale behind it. Recently, they developed a new technique called "Replica Model Predictive Control" that computes control quantities quickly and accurately, and validated the feasibility of using these in a control system at the large-scale boiler plant of Mitsui Chemicals Osaka, confirming that this technology can be safely and efficiently utilized for startup.

[Features of the Technology]

1. Reduces operation costs during unsteady states of chemical plants.

With this technology, efficient operational maneuvers can be achieved without waste during unsteady state operations, such as startup operations and production volume changes, which have previously been difficult to auto-control and assist.

2. Safe operations can be confirmed in advance.

In large-scale plants during unsteady states, the complexity of operational results makes future predictions difficult, adding to the challenges of operation.
This technology utilizes the dynamic simulator of Omega Simulation, allowing an AI to learn operational methods through reinforcement learning, repetitively conducting large amounts of trial and error automatically. The combination of reinforcement learning and precise simulation enables it to respond to various situations.

During operation, the mirror plant (online dynamic simulator) and the learned AI collaborate online, updating future operational contents and operational predictions in real time. This allows confirmation of whether operations are safe by reviewing the predictions.

3. Can quickly respond to unexpected volatility.

In plant operations, unexpected volatility can occur due to heavy rain and fluctuations in raw materials, and it is also necessary to respond quickly to these.

This technology further utilizes simulation data and neural networks to replicate the behavior of the plant as a replica model, and by equipping model predictive control using that replica model, can quickly respond to unexpected volatility.

The four parties will further develop the results obtained from this demonstration experiment and will contribute to the operational efficiency of chemical plants by providing operation support technology using AI and simulators.

That's all.