High-frequency time sequence data effective transmission method of intelligent factory based on long sequence dual prediction and informer

Abstract

The invention discloses a long sequence dual-prediction and informer-based high-frequency time sequence data effective transmission method for an intelligent factory, and relates to the field of intelligent manufacturing, and the method comprises the steps: firstly, building a cloud-edge collaborative long sequence dual-prediction architecture, then deploying a trained long sequence prediction model at an edge gateway and a cloud server of the architecture, and finally constructing a long sequence prediction model; and finally, reducing the transmission quantity of high-frequency data on line by adopting a long-sequence dual-prediction method, and ensuring the precision of the data. According to the method, the structure of a traditional dual-prediction method is improved, and the reasoning frequency of the prediction model is reduced through long sequence prediction, so that the application frequency of the traditional method is greatly improved, and it is possible that the method is used for reducing the transmission quantity of high-frequency data needed in the intelligent manufacturing process. And meanwhile, the latest deep learning model informer is introduced and combined to solve the problems of gradient disappearance and sharp increase of model reasoning time caused by long sequence prediction, so that the transmission quantity reduction proportion and the application frequency of the long sequence dual-prediction method are further improved.

Description

technical field [0001] The invention relates to the field of intelligent manufacturing, in particular to an efficient transmission method of high-frequency time series data based on long-sequence double prediction and informer in an intelligent factory. Background technique [0002] Smart manufacturing aims to realize the digitization and intelligence of factories. In this process, a large number of intelligent applications are deployed on the cloud platform, and these applications need the support of a large amount of online time series data collected by the underlying sensor devices, such as online product quality inspection applications and real-time equipment operating status evaluation applications. These data usually include high-frequency data such as vibration, current, voltage, etc., which have the characteristics of online, high data frequency and large transmission volume. As the number of applications and sensors continues to grow, the volume of these high-frequ...

AI Technical Summary

Problems solved by technology

[0005] 1. The scheme of reducing the amount of online time series data transmission in the non-double prediction method cannot guarantee the data accuracy

[0006] 2. The non-deep learning double prediction method has low prediction accuracy and can only be applied to low-frequency simple data that changes slowly

Due to the serial structure of LSTM, when faced with long sequence data, there is a problem of gradient disappearance, which makes it difficult to capture long-distance information.

As a result, it is difficult to perform long-sequence prediction, and the low prediction accuracy makes the transmission volume reduction ratio low

[0009] 5. The LSTM used by the latest bi-prediction method is a serial structure, which seriously affects the reasoning speed of the model

The increase in the inference time of the forecasting model reduces the applicable frequency of the proposed long-sequence bi-prediction method

Images