Three-phase unbalance is a major factor causing faults in three-phase four-wire distribution network systems, which can result in a significant decline in power quality, reduced power conversion efficiency, and other severe consequences. In this study, we present an early warning control scheme for detecting three-phase unbalance in the intelligent distribution sensing terminal of the three-phase four-wire distribution network. The following steps are undertaken. (1) The current sensor module of the intelligent distribution substation sensing terminal realizes the three-phase unbalance calculation of the point current data of the power grid and obtains a time-series dataset of current unbalance rates. (2) The parameters of a long short-term memory (LSTM) network are optimized using the quantum particle swarm optimization (QPSO) algorithm to determine the optimal network layer weights and thresholds during the training of the LSTM network. (3) A QPSO-LSTM-based time-series prediction model is developed to predict the current balance state. The accuracy and feasibility of the model are validated using a time-series dataset of current unbalance rates. (4) The aforementioned steps are integrated to design an early warning control scheme for three-phase unbalance in three-phase four-wire power distribution systems. This comprehensive early warning system enables the early detection and control of three-phase balancing states in the distribution system through the time-series prediction of the current unbalance rate. It facilitates the rotation or replacement of equipment that may disrupt the system’s balance, such as aging meters, and the timely detection and response to potential power system attacks. Although the overall early warning system requires a more stable and accurate power data acquisition technology to achieve the desired prediction, the proposed scheme provides valuable insights for controlling and compensating three-phase balancing and monitoring faults in three-phase four-wire circuits.