Research On Efficiency Prediction And Parameter Optimization Of Reciprocating Air Compressor Based On Digital Twin

Reciprocating compressors have been used in the industry field for more than 200 years.It is still the most preferred compressor equipment in the petrochemical industry,machinery construction,and many other fields because of its irreplaceable advantages such as high reliability,large compression ratio,and a wider range of varying operating parameters.Hence,it’s of great significance to further study the structure and working state of reciprocating compressors to improve its energy efficiency.In traditional investigations to improve reciprocating compressors’efficiency,researchers from the perspective of thermodynamics and operating dynamics,conduct a theoretical analysis of the compressor to explore the mechanisms and methods in the bid to improve its working efficiency.On the contrary,in this paper,the reciprocating air compressor is considered as a black box and by measuring,and sampling the actual physical quantities of the process,regression analysis is carried out.And then through mathematical modeling,a correlating relationship between the compressor’s input and output is established.This is also referred to as the construction of compressor digital twin.The correlating relationship between input and output variables can be derived quickly by the digital twin model.A digital twin model coupled with an automated control system can always attain constant optimization control of the compressor system.This implies that the compressor always runs in an optimal state of efficiency and hence achieves the aim of saving energy and reducing consumption.In this paper,the digital twin model is constructed based on a backpropagation(BP)neural network which is then optimized for its’ shortcomings.However,the traditional twin model based on BP neural network(BPNN)has lots of shortcomings,such as the need for a longer training time to establish a module,easily falling into the local optimal solution,and difficulty to achieve the global optimal solution.To solve these problems,a novel digital twin model based on the CIWOA-BPNN algorithm is put forward which determines the key indexes by principal component analysis method,and a CIWOA algorithm is introduced to improve the BPNN’s performance.The results show that the new CIWOA-BPNN twin model effectively avoids falling the local optimal problem.The main details of the research and results are as follows:1.Construct a test bench and then through laboratory experiments obtain 360 groups of experimental data(26 features)for statistical analysis.Collected data in the early stage of the process is checked for missing data and irregularities to ensure the integrity of the data.The principal component analysis(PCA)is used to determine the main factors affecting efficiency with reference to the rotational speed,clearance volume,cooling water flow rate,and displacement of the compressor.And the main influencing factors account for as high as 83.467%,which is very representative.2.Establish the digital twin model of the reciprocating air compressor based on BP neural network(BPNN).It’s however worthy to note that BP neural network often falls into the problem of local optimization which is mostly caused by improper settings of initial weight and threshold values.To solve this problem,a cubic map improved whale optimization algorithm(CIWOA)is proposed.This algorithm uses a Cubic Map chaos map to initialize the whale position.It assumes the initial best candidate in finding the solution of the target prey or the closest approximation to the optimal solution,and then the initial weight and threshold values are constantly optimized so that the problem of falling into local optimization can be solved quickly and effectively.To prove its accuracy and robustness,the GA-BPNN model,PSO-BPNN model,and WOA-BPNN model are established and respectively compared to the CIWOA-BPNN model.The CIWOA-BPNN model outperforms the other models and has the highest accuracy and strongest robustness.3.The improved CIWOA algorithm is used to optimize the BP neural network to obtain an optimal fitness of 0.0000364 after iterations reached the 11th generation.In comparison with other models established in this paper,the maximum and minimum relative error of the CIWOA-BPNN model is 0.0057 and 0.00005 respectively,and the correlation coefficient R is 0.99775,which is better than other models and has the highest prediction accuracy.The accuracy and robustness of the CIWOA-BPNN model as an efficiency prediction tool for reciprocating air compressors is verified.4.Perform efficiency predictions and parameter optimization for the digital twin model of the reciprocating air compressor to find the best pretraining and testing performance.For an efficient prediction,the optimal efficiency of a corresponding parameter can be obtained quickly when the value of the parameter variable is given.In parameter optimization,when given the compressor displacement,values of the corresponding rotational speed,clearance volume,and cooling water flow rates of the compressor can be quickly calculated under optimized efficiency.5.Finally create a visual GUI page that includes two modules.The first module is the reciprocating air compressor efficiency prediction module that calculates the corresponding efficiency of the reciprocating air compressor in real-time by the input of controllable parameters.The second module under optimal efficiency calculates specific values of rotational speed,clearance volume,and cooling water flow to finally achieve optimal control when the compressor displacement is given.