Research On The Hybrid Modeling Of Leaching Process With Ensemble Learning

With the rapid growth of the national economy, steady progressing of industrialization, the ores of high value and high grade are disappearing rapidly. Nowadays, the situation is becoming more and more severe. On the other hand, there are lots of deposits of low-grade non-ferrous metal resources in our country, so it is very important for our country to take the sustainable development strategy and use such resources effectively and economically.Compared with the traditional pyrometallurgy, hydrometallurgy is widely used in Metallurgical Industry with the advantage in dealing with these complex low-grade ores, and it brings little pollution to the environment. Leaching process is one of the central operation units in the hydrometallurgical processes and the basis of the next process, but, the control of leaching production process still remains off-line analysis, adjustments by experience and manual control.In order to achieve automatic control of leaching process, the online-measuring for leaching rate is quite imperative. In order to solve this problem, the thesis establishes the prediction model of leaching rate with the hybrid modeling based on the in-depth analysis of leaching process. Besides, considering that the model assessment is the premise and assurance of the model validity, so the model confidence is discussed; on this basis, according to the slow time-varying characteristic of the leaching process, model updating is performed. The main contents of this thesis include:(1) Based on the in-depth analysis of leaching process, the thesis builds the steady-state mechanism model of leaching process according to material balance mainly and lumped kinetic secondarily. The effectiveness of the model constructed is verified by simulations.(2) As the leaching process is so complex and that the mechanism model is built on a certain hypothesis and simplifications, the accuracy of leach-rate prediction is low with only the mechanism model. According to the problem given above, this thesis builds the model of leaching process by the hybrid modeling theory. It consists of two parts:the simplified mechanism model and data-driven model. The simplified mechanism model describes the basic characteristics of the leaching process, and the data-driven model compensates the error of not be modeled part. In order to improve the generalization ability of the model, the data model is built by ensemble learning method. At last, the simulation results verify the effectiveness of the hybrid model.(3) As model assessment is the premise and assurance for the model validity, and it also guides the updating of the model. So the model confidence is firstly determined based on the characteristic of ensemble learning. Meanwhile, according to the time-varying character of the leaching process, this thesis correct and compensate the hybrid model with the incremental learning method when its prediction accuracy can’t reach the expected result.