Situational Awareness and Fault Warning for Smart Grids Combined with Deep Learning Technology: Application of Digital Twin Technology and Long Short Term Memory Networks

In order to achieve effective perception of the power grid situation and accurate warning of operational faults, this study proposes a situation perception and fault warning method for smart grids based on deep learning technology. Firstly, using the digital twin smart grid platform as a carrier, build a smart grid digital twin situational awareness framework; Secondly, considering both dynamic and static security, intelligent grid situation evaluation indicators are selected; Then, comprehensively analyze the data of various indicators, evaluate the security situation of the power grid, and calculate the security situation assessment value of the power grid; Finally, a smart grid situational awareness model is built based on long short-term memory networks to achieve smart grid situational awareness and fault warning. A provincial-level smart grid big data information platform conducted experiments as the data source. After dividing the training and testing samples, 1000 iterations of learning were carried out to complete situational awareness and fault warning. The experiment was conducted to verify the accuracy, recall, F1 score, fault warning accuracy, fault command response time, and resource consumption of safety situation prediction results and actual values, as well as safety situation discrimination results. The experimental results show that the accuracy of this method for identifying the safety situation of smart grid operation is 98.72%, the recall rate is 98.95%, and the F1 score is 99.06%. This indicates that the comprehensive application performance of this method is good, and it can accurately and effectively perceive, predict, and analyze the safety situation of smart grid operation. At the same time, the maximum fault warning accuracy of this method is 99.82%, the minimum fault command response time is 0.083 s, and the minimum resource consumption is 118.57 MB, indicating that this method has a good power grid fault warning effect, which can accurately distinguish between normal operating conditions and critical states before faults and provide real-time and effective warnings.