Recently, due to the growth in machine learning and data mining, for scheduling applications in China’s industrial intelligence, we are quite fortunate to witness a paradigm of evolutionary scheduling via learning, which includes a new tool of evolutionary transfer optimization (ETO). As a new subset in ETO, single-objective to multi-objective/many-objective optimization (SMO) acts as a powerful, abstract and general framework with wide industrial applications like shop scheduling and vehicle routing. In this paper, we focus on the general mechanism of selection that selects or gathers elite and high potential solutions towards gathering/transferring strength from single-objective problems, or gathering/transferring storms of knowledge from solved tasks. Extensive studies in vehicle routing problems with time windows (VRPTW) on well-studied benchmarks validate the great universality of the SMO framework. Our investigations (1) contribute to a deep understanding of SMO, (2) enrich the classical and fundamental theory of building blocks for genetic algorithms and memetic algorithms, and (3) provide a completive and potential solution for VRPTW.
Single-objective to multi-objective/many-objective optimization (SMO) is a new paradigm in the evolutionary transfer optimization (ETO), since there are only “1 + 4” pioneering works on SMOs so far, that is, “1” is continuous and is firstly performed by Professors L. Feng and H.D. Wang, and “4” are firstly proposed by our group for discrete cases. As a new computational paradigm, theoretical insights into SMOs are relatively rare now. Therefore, we present a proposal on the fine brushworks of SMOs for theoretical advances here, which is based on a case study of a permutation flow shop scheduling problem (PFSP) in manufacturing systems via lenses of building blocks, transferring gaps, auxiliary task and asynchronous rhythms. The empirical studies on well-studied benchmarks enrich the rough strokes of SMOs and guide future designs and practices in ETO based manufacturing scheduling, and even ETO based evolutionary processes for engineering optimization in other cases.
13 March 2025 in IEEE Transactions on Instrumentation and Measurement
Silicon content prediction of hot metal is crucial for the stable and efficient operation of blast furnace ironmaking. As an emerging and promising data-driven modeling method, multiobjective evolutionary ensemble learning (MOEEL) has been successfully applied to silicon content prediction task. However, most existing MOEEL algorithms use an open-loop framework with two independent stages, i.e., base learner training and ensemble, which lacks information exchange between the two stages and limits their prediction performance. To address this issue, this paper proposes a closed-loop multiobjective evolutionary ensemble learning (CL-MOEEL) algorithm. In CL-MOEEL, multi-kernel support vector regression is used as the base learner, and the ensemble model is directly encoded as a solution in the population based on a variable-length encoding scheme. In this way, CL-MOEEL can simultaneously optimize ensemble model structure and base learner parameters, and utilize the ensemble results as feedback information to guide the training process of the base learners. In addition, a knowledge-guided selection method and two novel variable-length crossover operators are proposed to generate promising offspring solutions. Experimental results on both benchmark and practical industrial datasets show that the proposed method outperforms other powerful prediction methods. In the task of silicon content prediction, CL-MOEEL achieves an RMSE of 0.083, an R2 of 0.74, and a hit rate of 80.71%. Furthermore, the analysis of its complexity and feasibility highlights its strong potential for application in practical industrial processes.
As a new abstract computational model in evolutionary transfer optimization (ETO), single-objective to multi-objective optimization (SMO) is conducted at the macroscopic level rather than the intermediate level for specific algorithms or the microscopic level for specific operators; this method aims to develop systems with a profound grasp of evolutionary dynamic and learning mechanism similar to human intelligence via a “decomposition” style (in the abstract of the well-known “Transformer” article “Attention is All You Need”, they use “attention” instead). To the best of our knowledge, it is the first work of SMO for discrete cases because we extend our conference paper and inherit its originality status. In this paper, by implementing the abstract SMO in specialized memetic algorithms, key knowledge from single-objective problems/tasks to the multi-objective core problem/task can be transferred or “gathered” for permutation flow shop scheduling problems, which will reduce the notorious complexity in combinatorial spaces for multi-objective settings in a straight method; this is because single-objective tasks are easier to complete than their multi-objective versions. Extensive experimental studies and theoretical results on benchmarks (1) emphasize our decomposition root in mathematical programming, such as Lagrangian relaxation and column generation; (2) provide two “where to go” strategies for both SMO and ETO; and (3) contribute to the mission of building safe and beneficial artificial general intelligence for manufacturing via evolutionary computation.
Reheating furnace consumes a large amount of energy in the hot rolling stage of steel production, and has a significant impact on the product quality of hot rolling. To guarantee the stability of product quality, the operation optimization model of the reheating furnace is established. The operation optimization model aims to achieve optimal heating temperature settings for different slab specifications and different element contents, so that slab heating quality, heating temperature setting and slab heating time can be optimized simultaneously. To address this problem, an improved MOEA/D-DQN algorithm is developed according to the defined state and reward, in which the adaptive selection of differential mutation operators based on reinforcement learning is proposed to improve the search ability and robustness of the algorithm. Based on benchmark and practice, the experimental results show that MOEA/D-DQN algorithm can effectively solve the operation optimization problem of heating furnace.