Abstract: To address the challenges in fracturing stimulation such as large reservoir burial depth, high in-situ stress, and strong heterogeneity in the Block Anze of the Qinshui Basin, an intelligent cluster analysis of fracturing operation curves and anomaly diagnosis were conducted using unsupervised machine learning. Historical fracturing operation data from this block was systematically collected, and a multi-dimensional feature system was constructed by integrating time domain statistics, morphologic features, and symbolic shape features. spectral clustering and contrastive learning methods were adopted for feature extraction and intelligent clustering of operation curves to identify typical fracturing curve forms, reveal their corresponding fracture propagation modes, and quantitatively diagnose the parameter characteristics of abnormal well sections. The results show that there are five types of fracturing curves, namely low breakdown pressure and rapid propagation type, proppant placement difficulty type, stable pressure after high breakdown pressure type, stable pressure decline type, and stable pressure rise type. Abnormal sections exhibit the characteristics of three highs and two lows, i.e., high breakdown pressure, high operation pumping pressure, high fracture propagation pressure gradient, low medium-proppant ratio, and low average proppant concentration. The essence of the anomalies lies in increased near-wellbore friction under high-stress and insufficient fracture width, which lead to difficulties in proppant transportation. Accordingly, fracturing suggestions centered on "reducing friction, expanding fracture width, and optimizing proppant placement" were proposed, including pre-pad fluid ratio optimization (30%-50%), fine-proppant ratio control (35%-45%), and multi-slug friction reduction strategies, which can provide a reference for fracturing stimulation in similar coal reservoirs.