K-means结合RBF神经网络预测线性菲涅尔集热回路出口熔盐温度

线性菲涅尔集热回路出口熔盐温度控制具有扰动多、非线性及滞后的特点,出口熔盐温度的稳定控制可以极大的提高汽轮机组的发电效率,降低换热系统的调节难度、减少传储热设备的冷热冲击,提高系统使用寿命.针对传统数学模型预测线性菲涅尔集热回路出口熔盐温度精度低、计算复杂等问题,通过分析线性菲涅尔集热回路传热模型,确定影响集热回路出口...

Predicting molten salt temperature at the circuit outlet of Linear Fresnel heat collector using K-means combined with RBF neural network

Multiple perturbations, nonlinear and large hysteresis often occurred in the temperature variation of molten salt at the outlet of linear Fresnel heat collection loop in a solar thermal power station. The stable temperature control of molten salt at the outlet can greatly contribute to the generating efficiency of a turbine set, while easily regulating the heat transfer system, as well as the reduction of cold and heat shock in the heat transfer and storage equipment. However, there was a low precision and complicated calculation, when the traditional mathematical model was used to predict the molten salt temperature. In this study, a high-accuracy temperature prediction model was established using the Radial Basis Function (RBF) neural network with the K_means method. A heat transfer model of linear Fresnel heat collection loop was used to determine the main factors influencing the temperature at the outlet of molten salt in the heat collection loop. A means clustering algorithm was selected to analyze the input sample information, where the data center of each cluster was determined. The extension constant of the hidden layer basis function was determined via the cyclic adjustment in the training process using gradient descent. The network weight of the output in the basis function was obtained using the pseudo-inverse matrix technique. The small absolute average error and the minimum absolute error were obtained, when the number of hidden layer nodes was 30, after training the network with a large number of measured data. The 4-day measured data was selected to conduct the simulation test on the prediction performance of the network model. The maximum absolute error (MRERR) of the network prediction output was 121.23 ℃, and the maximum average absolute percentage error (MAPE) was 3.576 2E-4. The simulation results showed that the model could effectively predict the output temperature of molten salt at the outlet of the linear Fresnel heat collection loop. The prediction model was applied to the actual operation in the Dunhuang 50MW linear Fresnel photo thermal demonstration station. The findings can greatly guide the stable temperature control at the outlet of the heat collection loop in the station.