Abstract: Taking single-factor parameter fitting for boiler four-tube leakage characteristics into account, the non-stationary fluctuation properties of these parameters frequently lead to false alarms. Therefore, a multi-source data fusion-based early warning method for boiler four-tube leakage has been developed. Relationship functions are established between boiler load and key characteristic parameters-including inlet/outlet pressure, temperature, flow rate, wall temperature deviation, superheat, and oxygen content-at various heating surfaces. Multi-source data fusion is achieved by determining the distribution intervals of these parameters according to the boiler load state, and the fused multi-factor distribution intervals are adopted as the fitting parameters for leakage characteristics, thereby mitigating interference from non-stationary fluctuations of single-factor parameters. A binary leakage fault characterization method is employed to compute the fuzzy distance between the multi-source data fusion output and actual leakage faults. The linear variation state of fuzzy distance across consecutive time points is used as the basis for leakage early warning. On the test dataset, the proposed method resulted in only seven false alarm events under conditions of non-stationary single-factor fluctuation, indicating a significantly low false alarm rate.