As the construction of new power system gradually takes shape, the trend of source-load separation is becoming more and more obvious and its randomness is increasing, making the problem of voltage stability increasingly prominent. Under the new circumstances, the grid urgently needs a voltage stability assessment method with high accuracy, fast response speed and good extensibility. The static voltage stability assessment problem is defined as regression problem and artificial neural network is constructed to assess the problem online. Firstly, the training sample set is obtained by scenario simulation, power flow calculation and local voltage stability index calculation. Then the RReliefF method is used to sort the features and eliminate the attributes with low weight to improve the training efficiency. Then the mapping relationship between key features and voltage stability is obtained by artificial neural network training. Finally, taking the modified IEEE39-node system as an example, six groups of experiments are set and a simple linear weighting method is introduced to calculate a comprehensive evaluation index about the speed and accuracy of the model to verify that the proposed method has ideal modeling speed and high accuracy, and can meet the requirements of voltage stability assessment of power system under the new situation.