Queuing network model training method, queuing optimization method, equipment and medium

Abstract

The invention discloses a training method of a queuing network model, a queuing optimization method, equipment and a medium. The training method comprises the following steps: acquiring a real-time queuing network diagram; inputting the real-time queuing network graph as training data into a to-be-trained first graph neural network model; the first graph neural network model comprises a trained first model corresponding to the processor; in the training process, locking model parameters of the first model to be unchanged, and training the first graph neural network model based on a preset first loss function to obtain a first optimization network model corresponding to the encoder and a second optimization network model corresponding to the decoder; releasing model parameters of the first model, and continuing to train the first graph neural network model based on the first loss function to obtain a final queuing network model. According to the method, the trained processor is utilized, the encoder and the decoder are trained by utilizing the field data, training is performed again according to the field data, and relatively high prediction accuracy can be achieved by utilizing less data training.

Description

technical field [0001] The invention belongs to the field of queuing optimization processing, and in particular relates to a queuing network model training method, queuing optimization method, equipment and media. Background technique [0002] Generally, queuing theory algorithms are used to construct queuing mathematical models to deal with queuing problems. However, the assumptions in queuing theory on the probability distribution of arriving data (such as Poisson distribution) may be due to the long-tailed distribution of arriving data at the scene. The on-site conditions do not match, leading to calculation errors. Moreover, using manual analysis of on-site data and confirming the characteristics of the site, first, it cannot automatically make adjustments according to changes in the site conditions (such as: changes in the queuing sequence, adjustments to the number of queues); second, it cannot automatically establish a system that can be deployed in practice. A very ...

AI Technical Summary

Problems solved by technology

[0002] Generally, queuing theory algorithms are used to construct queuing mathematical models to deal with queuing problems. However, the assumptions in queuing theory on the probability distribution of arriving data (such as Poisson distribution) may be due to the long-tailed distribution of arriving data at the scene. On-site conditions do not match, leading to calculation errors

Moreover, using manual analysis of on-site data and confirming the characteristics of the site, first, it cannot automatically make adjustments according to changes in the site conditions (such as: changes in the queuing sequence, adjustments to the number of queues); second, it cannot automatically establish a system that can be deployed in practice. A very general mathematical model for various queuing scenarios

Images