At present, China has a developing natural gas market, and ensuring the security of gas supply is an issue of high concern. Gas supply reliability, the natural gas pipeline system's ability to satisfy the market demand, is determined by both supply side and demand side and is usually adopted by the researches to measure the security of gas supply. In the previous study, the demand side is usually simplified by using load duration curve (LDC) to describe the demand, which neglects the effect of demand side management. The simplification leads to the inaccurate and unreasonable assessment of the gas supply reliability, especially in high-demand situation. To overcome this deficiency and achieve a more reasonable result of gas supply reliability, this paper extends the previous study on demand side by proposing a novel method of management on natural gas demand side, and the effects of demand side management on gas supply reliability is analyzed. The management includes natural gas prediction models for different types of users, the user classification rule, and the demand adjustment model based on user classification. First, an autoregressive integrated moving average (ARIMA) model and a support vector machine (SVM) model are applied to predict the natural gas demand for different types of users, such as urban gas distributor (including residential customer, commercial customer, small industrial customer), power plant, large industrial customer, and compressed natural gas (CNG) station. Then, the user classification rule is built based on users' attribute and impact of supplied gas's interruption or reduction. Natural gas users are classified into four levels. (1) demand fully satisfied, (2) demand slightly reduced, (3) demand reduced, and (4) demand interrupted. The user classification rule also provides the demand reduction range of different users. Moreover, the optimization model of demand adjustment is built, and the objective of the model is to maximize the amount of gas supplied to each user based on the classification rule. The constraints of the model are determined by the classification rule, including the demand reduction range of different users. Finally, the improved method of gas supply reliability assessment is developed and is applied to the case study of our previous study derived from a realistic natural gas pipeline system operated by PetroChina to analyze the effects of demand side management on natural gas pipeline system's gas supply reliability.