Research On Black Carbon Emission Prediction Of Diesel Engine Based On Machine Learning And Swarm Intelligence

Black carbon emissions from ships raise the temperature of the earth’s atmosphere,accelerate the melting speed of Arctic glaciers,and pose a great threat to human health.At present,the International Maritime Organization(IMO)lists black carbon as one of the main pollutants emitted by ships,determines the nature and measurement methods of black carbon,and is calling on countries all over the world to carry out relevant research on black carbon emission,in order to realize the effective control of black carbon emission in a short time,so as to protect the earth’s environment.The measurement of black carbon emissions not only consume a lot of resources,but also brings errors to the measurement results due to various uncertainties in the measurement process,and in addition,the measurement results cannot meet the realtime requirements due to the long test time.Therefore,it is very important to develop a technique with high accuracy,fast response time and robustness for prediction of diesel engine black carbon emissions.Black carbon is one of the pollutants formed by the incomplete combustion of carbon-based fuel in the cylinder of diesel engine.The In-cylinder combustion process has an important impact on the generation of black carbon.Firstly,this paper analyzed the correlation between combustion characteristic parameters and diesel engine black carbon emission,and then took the combustion characteristic parameters as the input term to establish four prediction models of diesel engine black carbon emission based on machine learning,through the comprehensive comparison of the performance of each prediction model,the better models were selected.Finally,swarm intelligence algorithm is used to automatically optimize the hyper parameters of the better model,which improves the prediction accuracy of the model.The main research contents and conclusions of this paper are as follows:(1)The correlation between combustion characteristic parameters and black carbon emission of diesel engine.In this paper,firstly,bench tests were conducted on a diesel engine under different steady-state operating conditions,and five combustion characteristic parameters,namely maximum cylinder pressure,maximum pressure rise rate,maximum combustion heat release rate,maximum combustion heat release rate phase and heat release center phase,were extracted from the curves of in-cylinder pressure,combustion heat release rate and accumulated heat release,and then the relationship between each combustion characteristic parameter and black carbon emission concentration was analyzed.Finally,the correlation between characteristic parameters and black carbon emission concentration was quantified by mutual information.The results show that the normalized mutual information of each combustion characteristic parameter and black carbon emission concentration is greater than 0.9,and the adjusted mutual information is greater than 0.75,which proves that the correlation between the combustion characteristic parameters and black carbon emission concentration is strong and the combustion characteristic parameters can be used as the input term of the black carbon emission prediction model.(2)A study on the prediction of black carbon emissions of diesel engine based on machine learning.In this paper,four diesel engine black carbon emission prediction models,Lasso regression,Support Vector Machine(SVM),Extreme Gradient Boosting(XGB)and Artificial Neuron Network(ANN),were developed using combustion characteristic parameters as the input terms.Then in order to select a better model,the prediction performance,stability and training cost of each model were compared.The results show that:(1)The prediction results of linear model lasso on different data sets are poor,and the mean square error(MSE)of test set and training set are 1.8325 and1.6341 respectively.(2)SVM,XGB and Ann have achieved good prediction results.The MSE of test set and training set are 0.1483 and 0.0394;0.0585,0.0017;0.0831,0.0029 respectively.(3)ANN has strong prediction performance,but it has the defects of complex parameter adjustment process,easy over fitting,sensitive to the changes of input data and high training cost.(4)Although SVM,XGB and ANN can achieve good prediction results on the training set and test set,the average prediction relative errors of the three models for the prediction of the validation set are large,which are 17.83%,12.36% and 13.52% respectively.(3)Optimization of black carbon emission prediction model based on swarm intelligence algorithms.Aiming at the poor performance of diesel engine black carbon emission prediction model in the validation set,grey wolf optimization(GWO),Harris hawk optimization(HHO)and sparrow search algorithm(SSA)in swarm intelligence optimization algorithm were used to optimize the hyper parameters of SVM and XGB model respectively,and the prediction performance of the optimized model was verified.The results show that among the three swarm intelligence algorithms,HHO is the best algorithm for SVM optimization.The sum of MSE of the optimized SVM to the training set and the test set is 0.1414,which is 24.66% lower than that before optimization.The average prediction relative error of the optimized SVM to the validation set is 9.89%;The best algorithm for XGB optimization is GWO.The sum of MSE of the optimized XGB for training set and test set is 0.0487,which is 19.17%lower than that before optimization.The average prediction relative error of the optimized XGB for validation set is 8.61%.The research results of this paper can provide reference for the research and development of ship black carbon emission reduction and diesel engine performance optimization technology.