Power Failure Prediction System Based On Multi-layer Distributed Convolutional Neural Network

With the development of social economy,people’s demand for electricity quality is also increasing.The prediction of power system failure has gradually become a hot topic in power research.Along with the continuous development of artificial intelligence technology,traditional power system and artificial intelligence technology are constantly integrated,and a lot of technical means are generated,including expert system,artificial neural network,fuzzy logic diagnosis method and genetic diagnosis method.As an important branch of artificial intelligence,artificial neural network has been developing continuously in recent years.The emergence of deep learning marks the beginning of the craze of neural network.In this paper,multi-layer distributed convolution neural network is used to solve the problem of how to predict power system failure.By multilayer distributed convolution neural network design,a large regional power network system,can be divided into many small regional power network,the single after segmentation region corresponding to a single convolution of multi-layer distributed structure neural network,doing so can be used to reduce the difficulty of predicting complex power grid,and to reduce the amount of calculation of the neural network.A single convolution neural network is constructed by Alexnet,including convolution layer,pooling layer,full connection layer,etc.,and multiple convolution kernels are selected for testing.After the establishment of the model,a simple circuit was built through PSASP simulation power platform to simulate the resistance of R1,R2 and R3.In the circuit,current and voltage changes of the real circuit were simulated by modifying the resistance of R1,R2 and R3,so as to collect experimental data and train and test the neural network model.At the same time,this paper designs and realizes a complete power fault prediction system based on the development concept of software engineering,from demand analysis to summary design to detailed design.In addition,the multi-layer distributed convolution neural network model is adopted as a functional module of the system,including power information display,power information query,staff maintenance and other functional modules.The nearest distance between the predicted failure point and the repairman is redefined.Through the testing methods of black box and white box of software engineering,the functional modules of the system are tested,the prediction accuracy is tested with real regional power system data,and the data of analog circuit are compared to ensure the robustness of the whole system.Through this paper multilayer distributed convolution neural network model and forecast system design and implementation,we found that with the traditional BP(51%),SVM(83%),ANN(45%),in contrast.CNN provides a higher recognition rate of 93%,and the design of multi-layer distributed reduces the complexity of the model calculation,can make the power system fault information faster and more accurate was predicted.