The oil temperatures in wind turbine gearboxes are difficult to predict due to the strong nonlinearities. Multivariable correlations have been developed, but they are difficult to use due to the data redundancy between the correlation variables and the shortage of off-line training data for the artificial neural networks This paper presents a gearbox oil temperature prediction model based on a principal component analysis (PCA) and a dynamic neural network. The model uses online learning based on statistical process control (SPC). The PCA method deals with the data redundancy problem for the variables affecting the oil temperature. The nonlinear autoregressive with external input (NARX) dynamic neural network is then used to model the oil temperature. The SPC method analyzes the residual distribution to control the online learning behavior. Tests show that the method is stable and can accurately predict the oil temperature variations.
摘要 针对风电机组齿轮箱油温趋势预测中存在的信号非线性、多变量相关、各相关变量之间存在数据冗余等问题, 同时为了克服人工神经网络离线训练的不足, 该文提出了一种基于主成分分析 (principal component analysis, PCA) 和动态神经网络的齿轮箱油温趋势预测模型, 并结合统计过程控制 (statistical process control, SPC) 实现该模型在线学习能力。确定影响油温变化的相关变量集, 利用PCA消除相关变量间的数据冗余, 采用有外部输入的非线性自回归动态神经网络 (nonlinear autoregressive with external input, NARX) 对油温和相关变量集进行建模, 采用考虑残差分布规律的SPC方法控制模型在线学习行为。实际应用结果表明：该方法具有较高的稳定性和准确度, 能够有效实现油温趋势预测。