Load event detection is the most critical part of non-intrusive load monitoring systems. However, the general load event detection method is prone to missed detection, especially for slow load climb phenomenon, and is mostly limited to detecting the start time of the event, and does not involve the recording of the start and end process of the transient event. For this reason, this paper proposes a CUSUM transient event detection algorithm based on composite sliding window. Through the variable weight sliding window, the cumulative sample point deviation from the mean value is recorded, and the mutation start point and sampling point after the end of the mutation are recorded. The case validates the effectiveness of the algorithm. The algorithm not only can accurately detect the transient events with climbing characteristics that are easily overlooked by traditional algorithms, but also can effectively record the complete switching process, which helps to analyze and process transient events.