A method for generating time series based on generating antagonistic network

Abstract

The invention provides a method for generating time series based on generating antagonistic network. Based on the idea of generating antagonistic network, the invention designs a time series generating antagonistic network, and generates large-scale initial training data through limited time sequence data. The method includes designing a generator to generate data based on a long-term and short-term memory model LSTM, wherein the LSTM is used to learn potential temporal dependencies between the data for generating data with time dependencies that conform to the data distribution. A feature discriminator is designed based on scalable hypothesis testing to authenticate the quality of the generated data, the discriminator trains the LSTM-based generator by a policy gradient of reinforcement learning through a corresponding reward value of the data probability generated in each step of the temporal differential learning feedback generator, wherein the reward value is provided by the discriminator return value, thereby measuring the availability of the generated data.

Description

technical field [0001] The invention relates to the field of the Internet and the field of deep learning, in particular to a time series generation method based on a generative confrontation network. Background technique [0002] The time series generation method based on the generative confrontation network is mainly based on the idea of ​​the generative confrontation network. Through the identification feedback mechanism, the time series generation process is regarded as a continuous decision-making process to generate large-scale data sets. Among them, the discriminator extracts time series features and evaluates the importance of each feature to the sequence, and measures the quality of generated data by training real samples and generated samples. The discriminator learns the reward value corresponding to the probability of data generated in each step of the feedback generator through time difference learning, and the LSTM-based generator is trained by the policy gradie...

AI Technical Summary

Problems solved by technology

Feature discriminators based on scalable hypothesis testing designs have not been used so far, and this calculation of the quality of generated data relies on comparing features extracted from real data.

Images