Abstract: Stochastic configuration network (SCN) is a universal approximator which can be automatically and quickly constructed under the supervision mechanism with inequality constraint. It has potential advantages in the field of large data modeling. In order to enhance the accuracy and stability of the model prediction, a stochastic configuration network model with L2 norm regularization (L2-SCN) based on the classical SCN is proposed to improve the algebraic properties of output weighted least squares analytical solutions and avoid the structural risk of the model overfitting. For the ball mill load operation status recognition under a wide range of non-stationary operating conditions, L2-SCN method was used to identify the ball mill load operating conditions. The experiment results on the ball mill show that the proposed L2-SCN model has relative advantages in terms of accuracy and stability compared with the classic SCN model and the random vector functional link network (RVFL).

Key words: Randomized neural network, stochastic configuration networks, regularization, ball mill, condition recognition

摘要： 随机配置网络（Stochastic Configuration Network, SCN）在不等式约束监督机制下能够自动快速构建万能逼近器，在大数据建模领域具有潜在优势。为了增强模型的准确性和稳定性，在经典随机配置网络基础上提出带有L2范数正则化的随机配置网络，改善输出权重最小二乘解析解的代数属性，避免模型过拟合的结构风险。针对大范围非平稳操作工况下球磨机负荷运行状态识别问题，使用正则化随机配置网络构建球磨机负荷运行工况识别模型。球磨机实验结果表明，所提出的正则项SCN模型在识别准确性和模型性能稳定性方面相对经典SCN和RVFL模型，具有相对优势。

关键词: 随机神经网络, 随机配置网络, 正则化, 球磨机, 工况识别