Abstract:We present an automated framework that integrates rectified linear unit activated artificial neural network (ReLU-ANN) and mixed-integer linear programming (MILP) to enable efficient operational-level optimization of complex chemical processes. Initially, data is generated through rigorous simulations to pre-train surrogate models based on ReLU-ANN (classification and regression), and subsequently, MILP is employed for optimization by linearly formulating these models. This novel framework efficiently handles complex convergence constraints through a classification neural network which will be used for high-throughput screening data for regression, while simultaneously implementing an 'optimizing while learning' strategy. By iteratively updating the neural network based on optimization feedback, our approach streamlines the optimization process and ensure the feasibility of optimum solution. To demonstrate the versatility and robustness of our proposed framework, we examine three representative chemical processes: extractive distillation, organic Rankine cycle, and methanol synthesis. Our results reveal the framework's potential in enhancing optimization effect while concurrently reducing computational time, surpassing the capabilities of typical optimization algorithms. As for the three processes, optimization effectiveness improved by 10.11%, 28.69%, and 5.45%, respectively, while execution time were reduced by 71.71%, 54.49%, and 59.38%. This notable enhancement in optimization efficiency stems from a substantial reduction in costly while ineffective objective function evaluations. By seamless integration of ReLU-ANN and MILP, our proposed framework holds promise for improving the optimization of complex chemical processes, yielding superior results within significantly reduced timeframes compared to traditional approaches.

Real-Time Machine-Learning-Based Optimization Using Input Convex LSTM

Gradient information-based strategy for real time optimization and control integration

Hybrid-Model-Based Intelligent Optimization of Ironmaking Process

An Efficient and Flexible Learning Framework for Dynamic Power and Thermal Co-Management

Input Convex Lipschitz RNN: A Fast and Robust Approach for Engineering Tasks

A discrete-time neural network for optimization problems with hybrid constraints.

Identification-based Real-Time Optimization and Its Application to Power Plants

An optimization-oriented modeling approach using input convex neural networks and its application on optimal chiller loading

Predictor-Based Data-Driven Model-Free Adaptive Predictive Control of Power Converters Using Machine Learning

LSTM-based energy management algorithm for a vehicle power-split hybrid powertrain

Deep-learning modeling and control optimization framework for intelligent thermal power plants: A practice on superheated steam temperature

Development of novel dynamic machine learning-based optimization of a coal-fired power plant

Real-Time Optimization of Power Plant Boiler Based on Pattern Matching and Neural Networks

Linearization of ReLU Activation Function for Neural Network-Embedded Optimization:Optimal Day-Ahead Energy Scheduling

Optimizing urban energy management: A strategic examination of smart grids and policy regulations

Incorporating deep learning of load predictions to enhance the optimal active energy management of combined cooling, heating and power system

Intelligent optimization method for real-time decision-making in laminated cooling configurations through reinforcement learning

Automatic Remaining Useful Life Estimation Framework with Embedded Convolutional LSTM as the Backbone

Adaptive compensation using long short‐term memory networks for improved control performance in real‐time hybrid simulation

Accelerating operation optimization of complex chemical processes: A novel framework integrating artificial neural network and mixed-integer linear programming

Advanced hybrid LSTM-transformer architecture for real-time multi-task prediction in engineering systems