Digital energy meters often appear error phenomenon in the actual working conditions such as frequency fluctuations, harmonics, input noise and other factors. In order to study the measurement performance of the digital energy meter under the actual working conditions, a calibration method based on the actual working condition is presented. A condition recovery device is developed. A Blackman discrete Fourier transform (DFT) + adaptive linear (Adaline) neural network algorithm is proposed. The calculation of standard electrical energy is achieved. The digital energy meter calibration method and the wattage-second method are compared and analyzed. The results of error analysis show that the digital energy meter calibration method based on the actual working conditions and the wattage-second method can be used to verify the digital energy meter. But the test results of the former are smaller and more stable, and can provide reference for the performance evaluation of the energy meter under complex conditions.