制冷站双目标权重自适应非线性预测控制

针对传统制冷站控制系统易产生振荡, 且无法实现系统性能整体优化的问题, 本文提出一种制冷站非线性 预测控制策略, 优化目标函数设计为满足建筑冷量需求的同时, 尽可能提高系统整体能效. 为解决上述两个优化目 标之间的矛盾关系, 本文采用模糊逻辑设计了优化目标权重自适应模块, 实时求取权重因子最优解; 针对非线性系 统在线优化求解困难问题, 本文提出了基于神经网络的非线性滚动优化算法, 采用神经网络作为反馈优化控制器, 并将系统优化目标函数作为在线寻优性能指标, 结合Euler-Lagrange方法和随机梯度下降法对控制器权值和阈值进 行在线寻优, 算法计算量小, 占用存储空间适中, 便于采用低成本的现场控制器实现制冷站预测控制. 仿真实验结果 表明, 本文所提出的预测控制策略与PID控制相比, 在未加入优化目标函数权重自适应模块情况下, 系统平均能效 比提高约32.5%; 进行优化目标函数权重自适应寻优后, 系统平均能效提高约39.43%.

Bi-objective weighted adaptive nonlinear predictive control for air-conditioning refrigeration

In response to the problem that the traditional air-conditioning refrigeration control system is prone to oscillation and cannot achieve overall system performance optimization, this paper proposed a nonlinear predictive control strategy for air-conditioning refrigeration. The optimization objective function was designed to meet the building cooling demand while improving the overall energy efficiency of the system as much as possible. To solve the contradictory relationship between the above two optimization objectives, an optimization objective weight adaptive module is designed using fuzzy logic to find the weight factor optimal solution in real time. In order to solve the difficult problem of online optimization of nonlinear systems, this paper proposed a nonlinear rolling optimization algorithm based on neural network, using neural network as the feedback optimization controller, and using the system optimization objective function as the online optimization performance index, combining Euler-Lagrange method and stochastic gradient descent method for online optimization of controller weights and thresholds. The algorithm is computationally small, occupies moderate storage space, and facilitates the use of low-cost field controllers for predictive control of air-conditioning refrigeration. The simulation experimental results show that the predictive control strategy proposed in this paper improves the average energy efficiency ratio of the system by about 32.5% compared with the proportional-integral-derivative (PID) control without the addition of the optimal objective function weight adaptive module; after performing the optimal objective function weight adaptive optimization search, the average energy efficiency of the system improves by about 39.43%.