电冶金过程柔性负荷调度动态多目标闭环优化方法研究

Flexible load scheduling is to regulate or transfer electrical load on one’s own initiative for achieving interaction and balance between demand side and supply side, which is a key technology for constructing smart grid. In electrometallurgy process, it need consider two conflict objectives such as lowering power demand and improving production while scheduling flexible load. It is difficult to modeling the relation between electrical load and production index, and the relation is under the impact of minute- and second-scale changed working condition. The load scheduling problem brings challenges for the current optimization approaches. Inspired by the feedback idea in control theory, we will design a double-closed-loop optimization structure that applies to the two-scale changed process, and propose a real-time multi-objective memetic particle swarm optimization (PSO) based on a heuristic strategy. The strategy integrates case-based reasoning and online statistical learning to couple problem features and optimization knowledge on the design of PSO, which can improve its convergence performance. We will establish a set of fitness function models with multilevel accuracy and complexity, and then design an evolutionary selection approach of the models for reducing the evaluation time of fitness value. A series of simulation and field experiments are developed successively to verify the proposed approach. This research will not only provide effective theory for solving the flexible load scheduling problem in electrometallurgy process, but also bring considerable benefit for energy conservation and emission reduction.

柔性负荷调度是在需求侧主动调节或转移用电负荷，实现电网供需两侧的互动与平衡，是构建智能电网的基础技术。电冶金过程柔性负荷调度涉及降低电能需量与提高产量两个相冲突的目标，电力负荷与产量指标之间特性难以采用机理分析建立数学模型，且特性受到工况分钟与秒两种尺度动态变化的影响，对现有的优化方法提出了挑战。本项目借鉴控制中的反馈思想，设计适用于两尺度动态变化过程的双闭环优化结构，并提出基于超启发式策略的实时多目标Memetic粒子群优化算法。研究基于案例推理与在线统计学习相融合的超启发式策略，将问题特征与历史优化知识结合到算法中以提高寻优效率；建立具有多级精度的适应度函数模型，并研究多级精度模型的协同使用与演化选择机制，以缩短粒子群适应值评估时间，提高在线优化的实时性。对所提方法进行仿真实验与应用验证研究。本项研究不仅为电冶金过程负荷调度问题提供有效的理论方法，而且在节能减排方面具有显著的社会意义。

围绕电冶金过程负荷调度动态多目标闭环优化问题开展了研究。研究内容主要包括基于进化策略与机器学习算法的电冶金过程负荷调度模型建模与智能优化。提出了基于置信规则推理的电炉工况预测方法，实验表明工况预测准确率达到95%以上，提出了基于长短期记忆网络（LSTM）的电炉负荷需量动态预测方法，工业实验表明该方法MAE为5.72%、RMSE为7.8%和MAPE为2.996%，具有较高的预测精度。提出了电冶金过程柔性负荷调度实时多目标优化方法，该方法将基于案例推理与机器学习相融合，用来提高传统进化算法或群智能算法的寻优效率。在上述研究基础上，开发出一套电炉柔性负荷调度优化软件原型系统，进行了工业应用实验，结果表明所提方法比人工决策获得的群炉单吨电耗指标降低5%左右。本项研究不仅为电冶金过程负荷调度问题提供有效的理论方法，而且在节能减排方面具有显著的社会意义与经济价值。