Abstract: The problem of three-phase imbalance in distribution networks, caused by the random uncertainty in the operation of the single-phase equipment such as household photovoltaic power generation and residential power, becomes increasingly serious.Aiming at this problem, a method for stochastic modeling of load in low-voltage distribution networks driven by electricity consumption data is proposed. In this method, the Gaussian mixture distribution and the expectation maximization algorithms are used to describe the state of typical residential electricity consumption behaviors. The cloud states are divided through the establishment of cloud-cover level indices, and the power-output behavior states of the photovoltaic clusters are described with the Beta distribution. Markov chain is used to mine the historical operation data of the single-phase equipment cluster, and the transition matrix of the system operation states at each time is established. The sequential Monte-Carlo simulation method is used to sample the duration time of the behavior states, and a stochastic model of system operation states is established. Considering the structural characteristics and the stochastic model of the low-voltage distribution network, the Latin hypercube sampling is used to sample the node information of the distribution network. The power flow calculation is carried out based on the node-branch correlation matrix, and the assessment of three-phase unbalance problems is implemented. Finally, a distribution network model is adopted as a test simulation system to verify the rationality and effectiveness of the proposed method.