Fast Calculation Of Reliability Based On Deep Learning

Power system reliability is an important indicator to measure the level of the power system.How to achieve rapid and accurate reliability assessment of complex power grids has always been an enduring research topic in the field of power systems.However,with the expansion of the power grid,the performance bottleneck of traditional reliability evaluation algorithms has gradually emerged,and further improving the speed of reliability evaluation has become an urgent problem to be solved.This paper researches reliability evaluation methods based on deep learning and improves the reliability evaluation process of power systems by designing and optimizing deep network models.The main work of this paper is as follows:Firstly,this paper introduces the common methods and basic processes of power system reliability evaluation.Analytical method can accurately solve the reliability index,but it is not suitable for large-scale systems;Monte Carlo simulation method performance is less affected by the system scale,but the efficiency is low in the system with higher reliability.Therefore,this paper finds out the performance loss points in reliability evaluation,and studies the fast calculation method of load shedding based on deep learning,and launches the work of reliability evaluation.Secondly,for different application directions,two improved networks,FRCNN and UCNN,are designed based on neural networks.FRCNN is optimized for speed and is suitable for scenes with high-speed requirements;UCNN is optimized for accuracy to ensure the accuracy of the output to the greatest extent.For different models,optimization is carried out in terms of the loss function,optimization algorithm,hyperparameters,etc.,and a load shedding calculation process based on an improved network is proposed.With a lower misalignment rate,it has a shorter time consumption than traditional algorithms.Thirdly,according to the characteristics of the power system,a data set construction method for reliability evaluation is proposed.By comparing the influence of different failure orders on the system reliability,the hierarchical sorting method is introduced to construct,which effectively avoids the low density and the data set.High time-consuming construction process.At the same time,the random first three permutation sampling algorithms and the n-ary index random method are proposed to broaden the possible states of the data set and reduce the complexity of sampling time,thereby providing effective support for model training.Finally,a reliability evaluation model based on deep learning is proposed.Its network structure is replaceable.Both FRCNN and UCNN can complete the reliability evaluation task.The selection of the highest failure in the data set was analyzed in detail,the data set was constructed reasonably,and UCNN was used as the core module to evaluate the reliability of the IEEE-RTS79 system,and the reliability evaluation results and performance were compared with other algorithms.The results show that the evaluation model proposed in this paper is fast and effective.