Summary: modelling and optimization are central issues in engineering technology research, forecasting is an important prerequisite for scientific decision-making and planning, and predicting reliability is often an important indicator of technological success, but forecasting is a weak link in all technologies, as is the case for multi-causal areas. With the development of the national economy, in particular new energy needs and materials, multi-catalytic science and technology face new challenges in seeking new and efficient catalysts in the economic, safety, multifunctional and other areas, as well as a comprehensive examination of their performance in the context of the chemical process. Machine learning based on historical data and data-deep excavations have become important issues that need to be addressed urgently in the field of chemical informatics. The three key issues that are difficult and urgently needed to be addressed in the field of catalysts: the dynamic relationship model of catalysts, the active relationship model of catalysts, and the optimal design of catalysts, which are of great relevance for the characterization of the subject of catalysts, as well as for the control, optimization, simulation and so on. Work on multi-phase catalytic modelling and optimization, in combination with confusion and support vectors, is based on three main aspects of the study: model predictions for self-adaptation of mixed particle constellations and support for vector integration, design frameworks based on self-adaptation of mixed particle constellations and optimization of support vector machines, and time series model predictions for phase space re-engineering and support vector combinations. The research content of this paper is a cross-cutting field in disciplines such as information science, automation science and chemistry. The main tasks include: (1) the introduction of a projection algorithm based on self-adaptation of mixed particle constellations and support for vector combinations (adaptive chaos particle swarm optimisation-support vector review, acpso-svr), the introduction of the acpso inspiration-mapping mechanism for automatic selection of svr models over parameters, with a clear advantage over web-based search algorithms in the event of a significant change in the range of over-parametric values. The test of selecting the forest fires standard dataset in the uci machine learning database showed that the method had a high level of precision and good generalization, which was an effective way of addressing multiple variable regression predictions. The modeling methods presented were used for cu-zn-al-zr synthesis catalyst modelling and, in the absence of a reaction dynamics model, the catalyst component and motor model were obtained with good predictive results. (2) to propose an optimal design framework based on a combination of multi-target mixed particle constellation optimization and support vector machines. The multi-stage catalytic response is multi-phased and highly relevant, and it is very difficult to obtain catalyst performance indicators, using the trained svr model as an approximation model to optimize adaptation in the design framework, and seeking the best catalyst with global catalytic capabilities by optimizing the space for input variables through multi-target acpso algorithms. The application of this strategy to the development of the cu-zn-al-zr synthetic catalyst for diaether has shown that the performance indicators for two new sets of catalysts, given the best design method of acpso-svr, differ very little from the experimental test values, reducing the time spent on the development of catalysts, saving money and time consumption, and becoming a viable and effective new method of laboratory design catalyst. (3) a non-linear time-series prediction modelling method based on phase-space remodelling and supporting vector machine combinations (psr-svr). In response to the fact that complex mechanisms and activity of multiple phase catalysts in non-decided conditions are influenced by a number of factors, the acquisition of time-series data for the catalyst failure process is very limited and reduces modelling efficiency and predictive accuracy, a high-dimensional spatially recasting of data represents a way to re-construct the lost data series to achieve the regularity of the rewriting of the data and to explore the inherent complex nature of the catalyst failure. The modeling method was applied to modelling the failure process of the cu-si-al carbonate synthetic catalyst in the oxidation of methanol. The simulation showed that the predicted error of the catalyst failure model was within satisfactory range and that the given co2-time yield prediction could provide effective information for the correct design and operation of the reactor and for optimizing the response process。
