Abstract:

The extreme learning machine (ELM) problem can not quickly and accurately predict heat rate. Combined with swarm intelligence optimization algorithm, an ameliorated symbiotic organisms search algorithm and extreme learning machine (ASOS-ELM) comprehensive modeling method is proposed. This method uses the ameliorated symbiotic organisms search (ASOS) algorithm to optimize the parameters of the ELM hidden layer activation function to obtain the optimal ELM model. Firstly, the initial heat rate prediction model is established with ELM, and the root mean square error (RMSE) of the output heat rate is used as the fitness value of the algorithm. Then the appropriate ELM parameters are found through the ASOS algorithm to obtain an accurate heat rate prediction model. The performance of the heat rate prediction is compared with the traditional ELM model, support vector regression (SVR) model optimized by the ASOS algorithm, ELM optimized by improved particle swarm optimization (PSO) and basic symbiotic organisms search algorithm (SOS). The results show that the ASOS-ELM model has a precise forecasting ability and rapid convergence speed when dealing with complex data models, which provides a new idea for modeling the heat rate of a steam turbine.

Key words: steam turbine, heat rate, algorithm, extreme learning machine, optimization, model

摘要：

针对极限学习机（ELM）不能准确地预测汽轮机热耗率的问题，结合群智能优化算法，提出一种改进的共生生物搜索算法和极限学习机（ASOS-ELM）综合建模的方法。该方法利用改进的共生生物搜索（ASOS）算法优化ELM隐层激活函数的参数，求得最优的ELM模型。再将ASOS-ELM模型应用到热耗率建模中，首先用ELM初始化热耗率预测模型，以输出热耗率的均方根误差（RMSE）作为算法的适应度值，然后通过ASOS算法找到合适的ELM参数，从而得到准确的热耗率预测模型。并将热耗率预测的结果与传统的ELM模型、ASOS算法优化支持向量回归（SVR）模型、改进的粒子群算法（PSO）和基本的共生生物搜索算法（SOS）优化的ELM作对比。结果表明，ASOS-ELM模型在处理复杂的数据模型中，具有精确的预测能力与快速的收敛速度，为汽轮机热耗率建模提供了新思路。

关键词: 汽轮机, 热耗率, 算法, 极限学习机, 优化, 模型