Large-scale new energy grid connection has reshaped the control and operational characteristics of the power system. The existing power system state estimation methods face problems such as difficulties in identifying new energy fluctuation data, low estimation accuracy, and slow estimation speed. To help overcome the shortcomings of existing methods, a new energy power system state estimation method based on a skip connection (SC)-deep neural network (DNN) and multi-source data fusion is proposed. First, an improved interpolation method based on bidirectional long short-term memory (BILSTM) prediction is used for multi-source data fusion. Then a joint spatiotemporal crossover mechanism and data identification technology of the BILSTM network are used to replace the traditional measurement mutation detection method, thereby better handling new energy fluctuation data. Finally, a state estimation model based on SC-DNN is established according to the original measurement data set, and the fitting advantage of the residual module and the speed advantage of the neural network are combined, so as to improve the accuracy and speed of state estimation. An example analysis based on the IEEE39 bus system and a real network in an area of ??Xinjiang shows that, compared with the traditional method, the proposed method can more accurately distinguish new source fluctuation data and bad data, and at the same time improve the accuracy and speed of state estimation.