Latency-aware edge flow computation reinforcement learning partitioning method

Abstract

The application discloses a time-delay-aware edge flow computation reinforcement learning partition method and belongs to the technical field of mobile edge computing. In order to solve the problem of optimizing the processing time delay of an edge flow processing system through partitioning, the application comprises the following steps: a state space of a data partition model is established, and state data of the current time of the data partition model is collected; a processing time delay prediction model is constructed and is used for constructing a reward of a data partition decision model; the data partition decision model is established, including designing the input and output of the data partition decision model, designing the reward and a training method, and designing a hot key optimization method; state data is generated based on the state space of the data partition model and is used for training the processing time delay prediction model and the data partition decision model, so that a trained data partition decision model is obtained, scoring is realized for each data partition decision through the state data, the relative size of the scores is used for selecting a downstream operator instance to which data should be allocated, and a data partition decision is completed.

Images