The performance of a straight waveguide phase modulator is influenced by insertion loss and half-wave voltage, which in turn can lead to the degradation of fiber optical current transformers (FOCTs). Therefore, elucidating the impact patterns of insertion loss and half-wave voltage on the system error of FOCTs is crucial for further enhancing the performance of phase modulators. Based on the analysis of the working principle of the direct-coupled waveguide phase modulator, the theoretical models for both insertion loss and half-wave voltage of the modulator are established. Experiments are conducted to observe the variations of insertion loss and half-wave voltage under different environmental conditions, and to ascertain the influence of these variations on the system error of FOCTs. A long short-term memory neural network is utilized for error compensation of insertion loss and half-wave voltage in the direct-coupled waveguide phase modulator. The results show that after error compensation, the output error of the system under variable temperature conditions is within 0.05% and meets the accuracy requirements of the 0.2-class current sensor.