(1.中南大學(xué) 信息科學(xué)與工程學(xué)院，長沙 410083；2. 長沙民政學(xué)院，長沙 410004；
3. 湖南人文科技學(xué)院 生命科學(xué)系，婁底 417000)
摘 要: 針對硅錳合金埋弧熔煉過程中爐渣成分檢測難的問題，提出一種基于自適應(yīng)差分進(jìn)化 (ADE) 優(yōu)化的約減最小二乘支持向量機(jī) (RLSSVM) 軟測量模型。該模型以硅錳合金熔煉過程的工況參數(shù)為實(shí)測數(shù)據(jù)集，首先通過斯密特正交變換獲取高維特征空間核矩陣的基，然后利用Direct Kernel PLS回歸計(jì)算得到約減最小二乘支持向量機(jī)軟測量模型，并以最小化訓(xùn)練樣本的均方差為目標(biāo)函數(shù)，用自適應(yīng)差分進(jìn)化算法優(yōu)化最小二乘支持向量機(jī)的核參數(shù)和正則化參數(shù)，將此模型應(yīng)用于30 MW硅錳合金埋弧冶煉過程爐渣成分測量。結(jié)果表明：ADE-RLSSVM模型測量值與實(shí)際值的最大相對誤差為7.3%，運(yùn)行時(shí)間為21 min。
關(guān)鍵字: 埋弧爐；爐渣成分；差分進(jìn)化；最小二乘支持向量機(jī)；斯密特正交化
silicon-manganese smelting process
（1. School of Information Science and Engineering, Central South University, Changsha 410083, China;
2. Changsha Social Work College, Changsha 410004, China;
3. Department of Life Science, Hunan Institute of Humanities Science and Technology, Loudi 417000）
Abstract:To overcome the difficulty that the slag composition cannot be effectively measured in silicon-manganese smelting process, a soft sensor model based on reduced least squares support vector machine (RLSSVM) was proposed, which was optimized by adaptive differential evolution (ADE) algorithm. Firstly, based on the measured data, the base vectors of kernel matrix can be gotten by Schmidt orthogonalization in the high dimensional feature space. Then, the direct kernel partial least squares regression (PLS) calculation was conducted to obtain the RLSSVM soft sensor model. Taking the minimum standard deviation of the training sample as the objective function, the adaptive differential evolution algorithm was used to optimize the kernel function parameters and regularization parameter of LSSVM. At last, applying this method to estimate the slag composition in a 30 MW submerged arc furnace, the results show that the maximum relative error of ADE-RLSSVM model is 7.3% and the computation time is 21 min.
Key words: submerged arc furnace; slag composition; differential evolution; least squares support vector machine; Schmidt orthogonalization
