Online control method based on reinforcement learning for thickener

Abstract

The invention provides an online control method based on reinforcement learning for a thickener, and relates to the field of mining. The method comprises the following steps: acquiring historical recorded data monitored in the production process; establishing a control model with a dual-network structure composed of a model network and an evaluation network, and performing training on the model network and the evaluation network by utilizing the acquired historical recorded data; and predicting a underflow concentration and a mud layer height at a next moment through the trained model network,estimating a cumulative cost value at the next moment through the trained evaluation network according to the predicted underflow concentration and mud layer height at the next moment, calculating acumulative cost value at the current moment according to the estimated cumulative cost value at the next moment, and determining optimal control action including a underflow pump speed and a flocculant pump speed at the current moment according to the obtained cumulative cost value at the current moment by utilizing a gradient descent iterative algorithm. The method provided by the invention can reduce time consumption and improve control precision.

Description

technical field [0001] The invention relates to the field of mining, in particular to an online control method for thickeners based on reinforcement learning. Background technique [0002] In complex process industrial scenarios such as metallurgy, the thickener is a large-scale sedimentation tool that is widely used. It can concentrate a low-concentration solid-liquid mixture to form a high-concentration mixture through gravity sedimentation, which can reduce water and concentrate. effect. [0003] In the actual production process, due to the complex operating mechanism of the thickener, it is difficult to establish a mathematical model. Most of the control algorithms are based on the artificially designed expert system or the rule library in the fuzzy controller manually, and assisted by traditional proportional integral control. Means to realize the control of underflow pump speed and flocculant pump speed. Such methods rely too much on human experience and lack adaptab...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to provide an online thickener control method based on reinforcement learning to solve the problems existing in the prior art that it is difficult to establish a mathematical model, and the thickener control method relies too much on manual experience and lacks self-adaptability

Images