摘要
变风量空调控制系统具有非线性和动态特性。目前,在VAV空调控制领域应用最广泛的神经网络是静态前馈Bp神经网络,而在多层前向Bp网络中引入特殊关联层,形成有"记忆"能力的Elman神经网络,可以映射系统的非线性和动态特性。其在网络训练算法中,采用自适应学习速率梯度下降反向传播算法,显著提高了网络的训练速率,有效抑制了网络陷入局部最小点。文中分别采用Bp神经网络与Elman神经网络建立模型,对VAV空调系统的少量参数的数据进行仿真预测,经比较分析,证明后者具有收敛速度快、预测精度高的特点。
VAV air-condition systems is of the character of nonlinear and dynamic. Nowadays, the neural network which is used the most extensively is static BP neural network. An Elman neural network, which has a special correlation layer is appended to hidden layer of Bp network, can map nonlinear and dynamic behaviors. In the training algorithm of the network, a back-propagation algorithm with adaptive learning speed and momentum gradient failing is used, which can obviously prevent the network to trap in local minimum. The model tested by actual data from VAV air-condition system is established by using both BP network and Elman neural network .By analyzing and comparing, the latter features quick convergence speed and high forecasting precision.
出处
《自动化与仪表》
北大核心
2009年第4期35-38,共4页
Automation & Instrumentation
基金
湖南省自然科学基金资助项目(02JJY203)
中南林业科技大学青年科学研究基金重点项目(07010A)
关键词
ELMAN神经网络
BP神经网络
VAV空调系统
Elman neural network
BP neural network
variable air volume（VAV） air-conditions system
