Abstract: operational states. However，monitored data is susceptible to environmental interference and equipment aging，leading to significant threshold fluctuations. This discrepancy between the monitored state and the actual operating state reduces warning accuracy. To address this，an intelligent fault warning method based on multivariate state estimation technique is proposed. First，principal component analysis is applied to reduce dimensionality and retain key equipment parameters，thereby decreasing inter-parameter correlation. Wavelet transform is then used to effectively filter high-frequency noise from sensor data，completing the data preprocessing. Next，the multi state estimation technique is employed to construct a memory matrix encompassing all possible normal operating states of the equipment. This matrix is used to optimize weight vectors and generate estimation vectors， enabling accurate assessment of the equipment’s operational status. Finally, based on the estimation vectors，similarity calculation and dynamic threshold setting are performed. Combined with iterative optimization，the warning threshold deviation is computed to achieve intelligent fault warning. Experimental results demonstrate that the proposed method exhibits strong data-processing capability. It can trigger warnings earlier during the initial phase of fault evolution (the first 20% of the warning period)，with a warning threshold deviation of only 0.18. This indicates a close match between the quantitative warning results and actual fault data，confirming the accuracy of the warning system.