Aiming at the error estimation of smart meters in the substation area of distribution network, an error estimation method of smart meters was proposed based on particle swarm optimized BP neural network. Firstly, this method established the error estimation model of intelligent electricity meter from data collection and data prediction preprocessing; secondly, aiming at the limitation of the traditional BP neural network hidden layer node number, it proposes to optimize the hidden layer node number by particle swarm optimization algorithm, and constructs the BP neural network structure based on the optimized hidden layer node number to train the training sample data, then the obtained BP neural network calculates the test sample data to get the error data of intelligent meter. Aiming at the problem of smart grid operation error estimation in a typical distribution network area, the method established in this paper is used to evaluate the operation error of smart meters. The simulation example shows that the established model can effectively evaluate the operation error of intelligent watt hour meter, effectively improve the evaluation accuracy, and provide a basis for the discrimination of leakage and theft.