Energy Consumption Prediction Strategy Of Workshop Air-conditioning Chiller Based On Improved Xgboost Feature Selection Algorithm

In recent years,carbon emissions in various countries have increased year by year,and the global climate problem has become increasingly severe,and carbon emissions are closely related to economic development.In this context,my country has put forward the goal of carbon peaking and carbon neutrality,which promises to achieve the goal of no increase in carbon emissions by 2030,and to achieve the neutralization of carbon emissions and energy conservation and emission reduction measures by 2060.Therefore,energy conservation and emission reduction are imperative.In my country,the carbon emission of industrial energy is huge.Industrial energy refers to the energy used in industrial production activities,including power,heating and cooling,etc.The energy consumption of air-conditioning system accounts for 40%-60%.,And the energy consumption of the chiller accounts for about 40% of the total energy consumption of the air conditioner.Grasping the energy consumption of the chiller is of great significance to the normal operation of the factory and the optimization of energy conservation.With the rapid development of artificial intelligence technology,more and more data-driven methods are applied in various fields.The establishment of data-driven models for chillers can achieve rapid and accurate prediction of energy consumption.Compared with traditional physical and statistical methods,Data-driven methods can achieve real-time response,accurate prediction,better generalization performance,and a wider range of applications.This article mainly explores the optimization effect of the improved Xgboost feature selection algorithm on the three chiller energy consumption prediction models,and gives the optimal choice from the two aspects of time and accuracy.First of all,this article collected energy consumption data and related characteristic variable data on the water chiller system of a factory in Ningbo.In order to improve the characteristic data quality and model prediction performance,this paper uses scatter plot method,3σ method and box plot method to detect outliers in the original data set,uses cross-validation fitting grid search to optimize model parameters,and then deploys improved The Xgboost feature selection algorithm obtains the optimal feature subset,and finally the feature subsets before and after optimization are substituted into the Xgboost,ANN and RNN models to predict and compare the prediction results.The research results show that compared with a single feature selection strategy,the optimal feature subset obtained by using the improved Xgboost feature selection algorithm has improved effects in the three energy consumption prediction models,and the energy consumption prediction values of the three models are beyond the real The sample points outside the error range of 10% decreased by 5.47%,4.33%,and 0.71%,respectively.The average absolute error(MAE)and mean square error(MSE)of the predictive evaluation index were reduced,and the coefficient of determination R2 was improved;At the same time,the running time of the three models has also dropped significantly,and the average running time has dropped by 11.87%,10.62%,and 18.11%respectively;however,the running time of the three models has a large gap,the Xgboost model has the shortest running time,and the RNN model requires less time than The Xgboost model is two orders of magnitude higher,and the time required for the ANN model is moderate;the RNN model has the best prediction performance,and its prediction accuracy is 29.54% and 10.21% higher than that of the Xgboost and ANN models,respectively.Therefore,it is necessary to make trade-offs based on time and accuracy requirements in engineering practice.