On-site control method for online machine learning

Abstract

The invention provides an on-site control method for online machine learning. The on-site control method comprises the following steps that 1), model training is carried out, a signal channel is keptas a signal receiving end-an on-site controller-an instruction sending end to obtain instructions sent by the on-site controller and data received by the on-site controller, the received data are usedas input values, and the sent instructions are used as output values to carry out multi-section fitting on an insertion model until the received data and the sent instructions can be fitted to at least one kernel function, then the next step is executed, and a linear kernel function only comprises a basic form of a linear function; 2), strategy adjustment is carried out; and 3), intervention control is carried out. According to the method, a multi-section and two-stage fitting mode is adopted, so that effective online learning can be provided, the investment of manpower and material resourcesby enterprises can be effectively reduced, a research and development period is shortened, the handover of a control right from a traditional on-site controller to a machine learning insertion controller can be completed smoothly, and remote parameter adjustment is facilitated through an additional remote communication mode.

Description

technical field [0001] The invention relates to a field control method of online machine learning. Background technique [0002] At present, the application of machine learning in industrial control is gradually increasing. However, the biggest problem that local enterprises generally encounter at present is the severe lack of data, which makes it difficult to complete the training of machine learning controllers. A compromise solution is to divide it into two phases , first collect data for a period of time, and complete the code of the machine learning controller at the same time, and then perform training based on the collected small amount of data to obtain a first-stage machine learning controller, put it into use, and continue to collect data for a long time during use. Then retrain the machine learning model based on all the collected data to obtain a second-stage machine learning controller, which is used for final control. However, with this method, firstly, the R&...

AI Technical Summary

Problems solved by technology

However, using this method, firstly, the R&D cycle is too long, and secondly, the investment in manpower and material resources is extremely high. For enterprises, it is better to directly hire operators

[0003] In order to solve the above problems, the company designed the following figure 1 Shown is a field control system based on time-sharing control handover of control rights. The field control system can guarantee from the hardware that the handover of control rights from traditional field controllers to controllers with machine learning as the control core can be a gradual replacement. However, the existing technology does not provide technical enlightenment on how to complete the transfer of control

Images