This paper presents a new strategy of supervisory control applied to heating room system that is able to adapt to different operating conditions, in order to reduce the energy consumption and guaranteed a good thermal comfort for the occupants. The aim is to design a multi-model and multi-controller supervisor by using neural networks with a view to overcome the limitations of using a single controller and to deal with nonlinearities, parameters uncertainties and the time constant of the heating system. The proposed methodology is based on three model/controller pairs, a monitoring signal generator and a switching logic. Three architectures of neural network architectures, namely multilayer perceptron neural networks, radial basis function and memory neuron networks are used to evaluate this control technique and to design the different controller/model pairs which correspond to the same work office and a specific isolation state. A comparative study is associated to this work in order to recognize the best neural network in terms of performance and simulation time. The results obtained for three scenarios addressed herein show the importance and the effectiveness of this method.

本文提出了一种适用于采暖室系统的新型监控策略，该策略能够适应不同的工作条件，以减少能耗并确保乘员良好的热舒适性。目的是通过使用神经网络来设计多模型和多控制器的监控器，以克服使用单个控制器的局限性并处理非线性，参数不确定性和加热系统的时间常数。所提出的方法基于三个模型/控制器对，一个监视信号发生器和一个开关逻辑。神经网络架构的三种架构，即多层感知器神经网络，径向基函数和记忆神经元网络用于评估该控制技术，并设计与同一工作办公室和特定隔离状态相对应的不同控制器/模型对。一项比较研究与这项工作相关，以便在性能和仿真时间方面识别出最佳的神经网络。本文针对三种情况获得的结果表明了该方法的重要性和有效性。