Power CPS intrusion detection system and method based on multi-target variable-length CNN discrete optimization

The invention discloses an electric power CPS intrusion detection system and method based on multi-target variable-length CNN discrete optimization. Preprocessed historical monitoring data of an electric power information physical system are used as an input data set, and variable-length discrete coding is performed on hyper-parameters such as the number of convolution modules for constructing a convolutional neural network model, a model topological structure, architecture parameters and batch size of each convolution module, a learning rate, an optimizer type, weight regularization and the like; and designing a multi-target discrete optimization method to perform multi-target parallel optimization on a CNN model architecture and parameters based on a variable-length convolution module by taking an index of the CNN model on a verification set and the number of model floating point operations as optimization targets, thereby obtaining a Pareto optimal CNN model considering model performance and model complexi