Hi Manik_Sharma
1. If you cannot ground the neutral, then please have an attenuation network on the neutral as well; it is not recommended to just ground it with a 1Mohm resistor. Also, when you don't ground neutral, you are subject to measurement errors when the line voltages are not balanced. You may want to refer this app note for details. http://www.analog.com/static/imported-files/application_notes/AN-1334.pdf
2. I wanted to clarify the point about earth potential. Is earth at 12V? Is it the ground of your power supply design and also the ADE7763 IC? The schematic has a single ground symbol but some are labelled as 'VG' while some others are denoted as 'Vgnd'; the remaining are not labelled. Can you clarify the grounding part of the system? And, is your power supply design, an isolated one?
3. One important point to consider is that the VRMS measurement in ADE7763 is in fact a mean absolute value measurement (Refer to page 24 in Rev .C datasheet). So, the expected error is within +0.5%, if your input signal doesn't have any harmonics and the amplitude remains higher than Full-scale/20 signal level (Measurement bandwidth: 140Hz). Can you check the FFT of the input signal for harmonic content? May be when you are switching loads, the harmonic content varies. An easy test would be to test the setup with accurate low voltage fundamental-only inputs, applied directly on the voltage pins, thus enabling you to verify the performance of the VRMS computation precisely. Also, I am assuming that you are allowing the VRMS results to settle down to a final value and averaging the VRMS results over a second or so, to report a measurement.
Regards
hmani