Having a better understanding of the customs of consumers’ electricity consumption and the accurate load forecasting of individual consumer are of great significance to demand response implement and high efficient power system operation. Firstly the load characteristics of large consumers are analyzed, and it is pointed out that the load profiles of large consumers have characteristics such as large in volume and wide in coverage, variant in load features, highly short-term correlative with historic load, notable in fluctuation, and without clear periodic characteristic. According to these characteristics, a wavelet de-nosing and decision tree based method for pattern extraction and load forecasting is proposed. The method mines data of consumer’s historical load and extracts their utilization patterns, then personalizes loads forecasting of consumers based on different utilization patterns. Case studies on 50 typical large consumers in a province of China show that the proposed method is superior to other forecasting methods in accuracy.