Research On Application Of MAS Technology For Control In Grinding Process

Grinding process is a very important process in the metallurgical industry. The main work of the process is collecting the original ore from mine, and then putting the original ore through the conveyor, ball mill, the slurry pool, whirlpool and other equipments to remove the useless or harmful impurities from the original ore. The process is making the ore's particles from large to small, achieving the ore to monomer dissociation, and processing the ore into line with requirements of industrial production which is meeting the particle size distribution of ore of raw material. The grinding process is very important, so making an effective and reasonable control to ensure the stable operation throughout the beneficiation process is also important. Grinding process is a complex process which has a large number of equipments consisting of multiple-input and multiple-output. The process contains more variables, with time-varying parameters, strong coupling, large hysteresis and nonlinearity and many other features, which caused great difficulty. Hence traditional mineral processing technology needs to be improved. With computer technology, detection technology, artificial intelligence algorithm applied to automation, to improve the automation of the production process, to reduce production costs and improve product quality, to reduce energy consumption, to get a better economies of scale and competitiveness of the market, but also to meet the national trend of energy saving. It is very practical and long-term Significance.This paper focuses on the control solution which combines Multi-Agent with adaptive neuron-fuzzy system for grinding process. Modeling process in the control don't make all the inputs and outputs information together, and use intelligent algorithms of the grinding process as control strategies, but build the different modeling processes according the corresponding relation between inputs and control output. Centralized control builds intelligent control algorithm model between input and output control, and its advantage is that the grinding process as a whole control. When the rules are more in control, more complex input and output conditions, the calculation will be significant increase, and the consumption of time and resources will increase, especially in some high demand for real-time environment needs to be done to further adjustment. The control in this article are still the whole grinding process as a whole, the overall control of the various parts are not separated. The difference is that the control carried out in the overall refinement, the control results according to different control information to respectively make the control scheme, while maintaining overall control step. In this paper, grinding process control for them in three different control output:pressure, level, and the ore step, the three outputs independently controlled, the establishment of output related to input and output between the control models to control. Grinding of the existing control rules are refined and classified the results of isolation of the input parameters of each control, at the same time control results according to different control rules have been classified, the classification of the message of these applications individually controlled ANFIS modeling.When the adaptive neuro-fuzzy inference system (ANFIS) models the control objects, they can take full advantage neural networks and fuzzy inference advantages. ANFIS is based on neural network fuzzy inference system to make use of existing knowledge and rules, and adjust these rules. In this paper, using back-propagation algorithm to adjust the premise parameters and consequent parameters, and the average training error and validation error are able to achieve reasonable accuracy, in line with the expected results.This article builds a self-learning program of grinding process which is based on blackboard model by using JADE and ANFIS. Using the advantages of multi-Agent on communication and collaboration, established the model which is based on the blackboard model. According to the intent and objectives' differences, Agents in blackboard model are divided into three categories:decision-making Agent, control Agent and the blackboard Agent. Blackboard Agent mainly provides the data exchange services. There are three control Agents in the model:control Agent for pressure, control Agent for level, and control Agent for mining step. And each of control Agent directly combines with ANFIS. The control Agent is mainly responsible for the implementation of the control process. The reason which is called self-learning is the use of the blackboard model of ANFIS with their own characteristics. Because the blackboard Agent can save data from ANFIS, and the results' control accuracy is higher. When the program starts, the accurate data is used by ANFIS training and studying, through such a cycle, the control process will be more stable, and the control information will be more accurate by using the accurate data for training. Finally, the blackboard model with the self-learning scheme for control in grinding process is based on the above technology.