Multi-rate Layered Operational Optimal Control Of Grinding Process

Grinding process is the prerequisite process of mineral processing technology,which aims to monomer dissociate useful minerals and gangue from ore to the maximum extent.Grinding process provides products that meet the requirements of particle size for the next beneficiation process.The operational state of grinding process directly determines the concentrate grade of mineral processing products,and affects the production indices such as mineral processing energy consumption level and metal recovery rate.It is very important to realize the operational optimal control of grinding process.Under the premise of considering the influence factors such as ball mill ore supply and classifier water supply,the system can work stably in the low energy consumption and high efficiency operational state.It is of great social and economic value to increase the production and efficiency of grinding process by making the grinding particle size and cyclic load meet the process requirements.The controlled objects of the operational optimal control of grinding process include the basic loop process with fast time scale and operational process with slow time scale.The dual-layer control structure of the basic loop layer and the operational layer is often used.The controlled object of the basic loop layer is to achieve fast and stable tracking control of basic loop.And the corresponding ore and water supply processes have fast time scale characteristics.The controlled object of operational layer is to control the two operational indices of grinding particle size and cyclic load within the expected range by optimizing the setpoints of the basic loop.And the corresponding ore grinding and selection processes have slow time scale characteristics.In the meantime,due to the different characteristics of various instruments,it is difficult to keep the same sampling and control periods in the basic loop layer.So the operational optimal control of grinding process is a multi-rate problem.In addition,the mathematical model of grinding process is difficult to be accurately constructed due to the continuous change of factors such as the frequent fluctuation of ore composition and properties,and the time-varying operational equipment.It is difficult to use the existing model-based control methods to solve the multi-rate control problem of the operational optimal control of grinding process with unknown model.This paper focuses on the grinding process and adopts data-driven technology to study the multi-rate layered operational optimal control of grinding process.The main work includes:1)Aiming at the characteristics of multi-rate and the difficulty in establishing accurate process model in the basic loop layer of grinding process,a data-driven basic loop layer control method is proposed by combining block lifting technique with Qlearning.Firstly,the block lifting technique is used to increase the inconsistent output sampling period and control refresh period of basic loop layer to a unified framework period.And the single-rate generalized controlled object is established.Then,an online updating controller based on Q-learning model-free method is designed to realize the tracking control of the basic loop setpoint.And the convergence of the algorithm is analyzed.2)Aiming at the different time scales of operational layer and basic loop layer of grinding process,and the difficulty in establishing the nonlinear model of operational layer,the iterative lifting technique and the actor-critic algorithm are combined to propose a data-driven control method for the operational layer of grinding process.Firstly,the iterative lifting technique is used to increase the period of basic loop layer to the period of operational layer.The closed-loop basic loop system is brought into the operational layer model.Then,the operational layer model is extended to obtain the generalized controlled object with the setpoint of loop layer as the input value and the operational index as the output value.Thus,the model-free method based on actor-critic neural network is adopted to realize the optimization of loop layer setpoint,and the relevant convergence proof is given.The proposed method is verified by simulation experiments on the self-developed grinding process simulation platform.The contrast experiments with the traditional algorithm show the superiority of the proposed method.