Three-phase unbalance refers to the inconsistency of three-phase current or voltage amplitude in a power system, which is a leading cause of power quality degradation, increased line loss rates, and transformer failures in distribution network systems. In this study, we propose a set of early warning methods for detecting three-phase unbalance states using data fusion from current sensors, current balance rate state analysis, and timing prediction. The proposed method utilizes the data collected from current sensors in a smart metering system to establish timing data for current unbalance rates through the calculation of unbalance degree and the coding of unbalance states. The parameters of a backpropagation (BP) neural network are optimized using a genetic algorithm (GA) to improve prediction accuracy by determining optimal values for neuron weights and thresholds during network training. Finally, a current balance state timing prediction model based on the GA-BP algorithm is established and validated using the collected data to verify its accuracy and feasibility. While the overall early warning system may require more precise current sensors to provide stable and accurate electrical energy data for prediction, the proposed method can achieve a balanced situational awareness of the three-phase power system and provide an effective decision-making basis for deep security defense, and take the necessary measures in a timely manner to respond to the problems encountered in the power system.