Online Monitoring Method for Operation Status of the thermal power generating units Based on Fusion of Limit Learning Machine and SCADA Data

Yongjun Fu¹Xuyang Zhao^1*Zhiqiang Yang²Yujiang Zhang³Yaping Qv³Baohua He¹Yuewu Yang¹

• ¹Power Control Center Of HUOLINHE Circular Economy, Huolinguole, China
• ²SPIC NEI MONGGOL Corporation, Tongliao, China
• ³SPIC NEI MONGGOL CORPORATION, ELCTRIC POWER BRANCH, Huolinguole, China

To quickly learn the complex operational patterns of thermal power generating units, this paper proposes an online monitoring method based on extreme learning machines (ELM) and SCADA data. First, health data from normal operations of the thermal power generating set are collected, and outliers unrelated to the unit's operating conditions are removed from the SCADA data, along with local abnormal points. Using this processed data, an extreme learning machine model is trained and constructed. The initial hidden layer output matrix and weights are obtained by adjusting the hidden layer output weights and training the model with the SCADA health dataset. During the updating phase, the model continuously learns from the input SCADA health data and updates the hidden layer output matrix and weights in real time until training is completed, enabling online monitoring of the thermal power generating set's operational state. The experimental results demonstrate that the proposed method can track the operating state of thermal power generating units in real time and detect abnormalities or faults. When the number of hidden layer nodes reaches 75, the gradient vanishing curve of the Tanh function initially lies below that of the RBF function but gradually rises and surpasses it. On the test set of SCADA dataset 5 for thermal power generating units, as the number of hidden layer nodes increases, the gradient vanishing curves of all activation functions exhibit an upward trend. The adopted extreme learning machine model demonstrates the best performance in predicting and classifying the unit's operating state, improving accuracy and showing higher efficiency.