基于并联形式的智能融合算法的燃气轮机仿真模型构建

Modeling and Simulation of a Gas Turbine Based on Parallel Intelligent Fusion Algorithm

为更好地实现机理模型与智能预测模型的优势互补,采用机理、BP神经网络和逐步聚类3种单一模型,分别构建了包括最优加权法、BP神经网络融合算法和多模式动态权重分配法在内的3套机理模型与融合算法相结合的智能融合模型。以辽宁某钢厂燃气轮机为研究对象,对比分析了3种智能融合模型与单一模型的仿真效果。结果表明:智能融合模型的仿真精度更高,均方根误差均在1%以内;当训练数据量较多时,智能融合模型中最优加权法的准确度比其他2种模型高,其决定系数高于0.97,平均误差为0.51%;融合算法在燃气轮机仿真模型构建的成功应用,为提高燃气轮机模型仿真精度和掌握变工况条件下的表现提供了很好的技术支持。

In order to better realize the combination advantage of mechanism model and intelligent prediction model, on the basis of three single models(i.e., mechanism, BP neural network and stepwise cluster models), three types of hybrid parallel intelligent fusion models composed of mechanism model and intelligent prediction model were firstly established by weight-optimization method, BP neural network fusion algorithm and multimode dynamic weight allocation method, respectively. Taking the gas turbine of a steel plant in Liaoning Province as the research object, the simulation effects of three intelligent fusion models and three single models were compared and analyzed. Results show that the accuracy of the intelligent fusion models is higher, and the root mean square errors were within 1%. When the amount of training data is large, among three intelligent fusion models, the weight-optimization method has the higher precision than other two models, its coefficient of determination is higher than 0.97 and the average error is 0.51%. The successful application of the fusion algorithm in the gas turbine simulation was capable of providing good technical support for improving the prediction accuracy of simulation model and estimating the performance of the gas turbine under variable working conditions.