Distributed asynchronous optimization method based on continuous convex approximation

Abstract

The invention relates to a distributed asynchronous optimization method based on continuous convex approximation and belongs to the field of network communication. The method comprises the following steps of: S1, constructing a problem model, namely selecting a global objective function of an actual problem; S2, initializing a local variable held by each node, and setting the maximum number of iterations; S3, constructing a directed strongly-connected unbalanced source network, and adding virtual nodes on the basis of the directed strongly-connected unbalanced source network to construct an augmented network; S4, setting a time delay threshold, the number of iterations and system parameters; S5, determining the relationship between an activated node and a time delay value associated with the activation node; S6, clearing out-of-date information in the system; S7, selecting an agent function, and setting proper step length and momentum parameters; and S8, enabling the activated node tocommunicate with the neighborhood or update by utilizing a time delay value variable, and enabling a non-activated node to keep the current variable value to enter the next iterative update until thethreshold value of the number of iterations is reached. The method has high robustness and fault tolerance for an asynchronous network, and the utilization efficiency of a communication link is improved.

Description

technical field [0001] The invention belongs to the field of network communication, relates to the technical field of distributed optimization of large-scale network communication data, in particular to a distributed asynchronous optimization method based on continuous convex approximation technology. Background technique [0002] In the early days, various industries were in the initial state of development, and the data to be processed was often small in size and low in dimensionality. Therefore, these data often only needed to be stored as a system, and then traditional centralized optimization algorithms could be used to efficiently process these data. However, with the rapid development of large-scale networks, data types, volumes, and dimensions increase, and data are often stored in multiple systems. Traditional centralized optimization algorithms have been difficult to meet the growing needs of the industry, so people began to focus on the design of distributed optim...

AI Technical Summary

Problems solved by technology

[0007] (1) The existing centralized optimization technology has problems such as single point failure and poor robustness. With the popularization of large-scale networks, it has been difficult to meet the development needs of various industries;

[0008] (2) The existing distributed synchronization optimization algorithm will cause a large amount of idle time in the communication link, resulting in a waste of network resources

At the same time, the requirement of clock synchronization is not suitable for the practical application of large-scale network;

[0009] (3) The existing asynchronous algorithms have slow convergence speed and can only deal with general unconstrained distributed optimization problems, and the research on distributed composite optimization problems is extremely scarce

Images