Research On Production Operation Analysis Method Of Chemical Equipment Based On Time Series Data

The chemical industry takes an important part in our country’s national economy,and its technical level affects industrial strength and people’s quality of life.The chemical industry provides our country’s main energy sources such as gasoline and diesel,and its products are also used as raw materials for agriculture,construction and material industries.The chemical industry not only pays a lot of resources such as fossil energy and electricity,but also emits a lot of air pollutants and water pollutants in the production process.Therefore,it is of great significance to promote the sustainable development of the chemical industry and the refinement and informatization construction of chemical equipment to enhance the strength of our country’s industrialization and achieve the sustainable development of our country’s industry.This paper studies and analyzes the historical time series data generated in the production process of chemical equipment taking catalytic cracking unit as example,proposes the energy efficiency analysis and forecasting method of chemical equipment.In this paper,the data envelopment analysis cross model(DEACM)based on GBDT(Gradient Boosting Decision Tree)(GBDTDEACM)method is proposed to analyze the energy efficiency of chemical equipment,improve the input-output structure for chemical equipment and provide theoretical guidance for improving chemical equipment’s energy efficiency.Then this paper proposes the Multi-Scale Temporal Long ShortTerm Memory(MST-LSTM)model and the MSPNet(Multi Step Predict Net)model by analyzing the time series data characteristics to achieve predicion of single-step and multi-step key indicators of chemical equipment accurately.The result provides strategic support for energy efficiency improvement and smooth operation of the device.The contents of this paper include:(1)Owing to the small fluctuation of the indicators of chemical equipment over a period of time,the traditional Data Envelopment Analysis(DEA)method cannot effectively distinguish the effective Decision Making Unit(DMU)and the invalid DMU in the energy efficiency analysis.Therefore,traditional DEA is impossible to provide effective guidance and optimization suggestions for chemical equipment.In view of the above problem,this paper proposes the GBDT-DEACM energy efficiency evaluation model.This paper analyzes the energy efficiency of the diesel production part of the catalytic cracking unit,and provides an optimization direction.The laboratory results show that the energy saving potential of the diesel production part of the catalytic cracking unit reaches 9.32%.(2)The energy efficiency analysis method is to analyze the historical state of the chemical equipment through historical time series data,but cannot realize the evaluation of the future state of the chemical equipment’s input and output based on the energy efficiency improvement strategy.Therefore,using appropriate modeling prediction method for chemical equipment modeling prediction not only provides data support for energy efficiency analysis of chemical equipment but also analyzes of chemical equipment’s production running state comprehensively.Due to traditional data-driven modeling cannot effectively mine the time-series features in the real production process of chemical equipment,this paper proposes a multi-time-scale neural network model the MST-LSTM to achieve accurate single-step prediction of chemical equipment.The MST-LSTM model captures the relationship between the relevant variables and the target variables in the long-term time series data and the autoregressive trend of the target variables in the short term through the structure of two neural networks,and adaptively adjusts the reference quantity of the information extracted by the MST-LSTM to the two neural networks through the gating mechanism.Finally,the Bayesian optimization method is used to adjust the time window size of the input time series data to further optimize the prediction ability of the MST-LSTM.Taking the input and output of the catalytic cracking unit and other key indicators of the catalytic cracking unit as the data set,the MST-LSTM has achieved an effective single-step prediction of the future state,and the accuracy rate has reached more than95%.Comparative experimental results show that the MST-LSTM performs best in regression metrics compared with other single-step prediction methods such as traditional Long Short-Term Memory(LSTM).The root mean square error(RMSE)of the MST-LSTM’s prediction can be reduced by up to 55%compared with traditional LSTM,which fully demonstrates the prediction performance of the MST-LSTM.(3)In the chemical equipment’s energy structure adjustment direction and the analysis of production status,the information support provided by singlestep forecast is limited,and it is impossible to evaluate the changes of indicators in the future.Therefore,it is meaningful to introduce multi-step forecast for chemical equipment.Aiming at the problem that the prediction deviation of traditional multi-step prediction method adds with the increase of prediction steps,the MSPNet is proposed for the operation analysis of chemical equipment.Taking the input and output data and other key indicators data sets of catalytic cracking units as examples,the experimental results indicated that the multi-step prediction method proposed in this paper can accurately achieve multi-step prediction,and the RMSE is as low as 0.006.In comparison with multi-step prediction models such as the Seq2 Seq,the prediction effect of the MSPNet is significantly improved,and the RMSE can be reduced by up to 87%.