Integrating multidimensional operational parameters for abnormal diagnosis in substations: A composite approach of non-uniform time series segmentation, trend information extraction, and symbolic representation

The abnormal operation of heating systems reduces efficiency and safety. Anomalies in the heating system have high-dimensional temporal features and interactions between multidimensional operating parameters. Traditional feature extraction methods often rely on manual experience and fail to accurately capture abnormal information. Therefore, this paper proposes an operating fault condition diagnosis method based on a symbolic representation of time series (SRTS) algorithm. It converts time series data, such as supply temperature, valve opening, instantaneous flow and heat, into symbolic representations. First, a small amount of abnormal data is labeled based on expert knowledge in the heating field. Time series rules are formed through trend segmentation, slope calculation, and threshold judgment. Then, thresholds and time rules are applied to the second step training set, generating pseudo-labeled data. Combining the pseudo-labeled data with the manually labeled data, the second training is conducted on the entire training set to develop the final diagnosis model. The model is evaluated on test set data from typical substations. The accuracy, precision, recall, and F1 score for diagnosing heat source failure conditions and electric valve fault conditions all exceed 95 %. For less frequent substations shutdown conditions, the accuracy and precision reach 91 %.