摘要
针对大孤山选矿厂磁选工艺过程的多指标强耦合、时变、非线性和大滞后等特点,使基于数学模型的常规预测方法难以应用问题,提出一种新型的权函数神经网络建立能源消耗预测模型。该模型网络拓扑结构只有输入输出两层,网络权值由传统的常数改为权函数。在权函数构造上,结合选矿厂实际生产过程中所提供的生产数据,根据数据样本间隔距离的大小,分别采用不同的函数作为网络的权函数。训练算法仿真实验表明,该算法计算量小,且建模误差为10-2数量级,取得很好的预测效果,从而克服了传统算法局部极小与收敛速度慢的问题。
The characteristics of multivariate strong coupling, time varying, nonlinear and long time-delay in the magnetic separation process of Dagushan Ore Dressing Plant, make it difficult to use the conventional methodologies of optimal control based on mathematical model. With simple network topology constituted by input layer and output layer only, the new neural network with weight functions is proposed. The weight is function instead of traditional constant. On constructing of weight function, according to the production data in actual production process of Ore Dressing Plant and the gap of these data, different interpolation functions are selected as the weight functions. Simulation examples show the good performance of this method that little calculation work, high calculation speed, with no local minimum and slow convergence problems. Model mentioned above has minor error and the better prediction effect is obtained.
出处
《吉林大学学报（信息科学版）》
CAS
2009年第1期73-78,共6页
Journal of Jilin University（Information Science Edition）
基金
国家科技支撑基金资助项目(2007BAE17B04)
关键词
神经网络
能耗预测
权函数
neural networks
energy consumption forecasting
weight function
