Research And Implementation Of Analysis Method For Influencing Factors Of Non-Oil Sales In Gas Stations

With the increasingly fierce competition in China’s refined oil retail market,in order to improve overall profits,oil companies need to dig out the factors that affect the sales of non-oil products in convenience stores in gas stations to increase non-oil product profits.This paper takes the nonoil sales data of a company as the research object,and researches and implements an analysis method for the influencing factors of sales.At present,among the related methods of influencing factor analysis,the algorithm based on gradient boosting decision tree is more suitable for processing data sets with large amount of data and high dimension,but there are still some problems:In response to the above problems,the main work and contributions of this paper are as follows:(1)When facing high-dimensional data sets,there will be an over-fitting problem,resulting in large errors in the analysis results.If the traditional feature screening method is used for feature screening of the data in this article,the sparse but important features will not be retained;(2)Hyperparameter optimization plays a crucial role in model performance,and manual adjustment requires relevant experience and consumes a considerable amount of time;In addition,in the quantitative analysis of influencing factors,such methods can only express the degree of influence of characteristics on the dependent variable,and cannot obtain whether the influence relationship is positive or negative.In response to the above problems,the main work and contributions of this paper are as follows:First,a quantitative analysis model of influencing factors is proposed.The model consists of an eigenvalue screening module and an eigenvalue information gain calculation module.The model aims at the shortcomings of the existing methods,and the main improvement points are:(1)An eigenvalue screening module based on adaptive kernel Lasso is designed.This module expands the eigenvalues into new features through the feature encoder,and then uses the adaptive kernel Lasso to filter the features with the granularity of the eigenvalues,reducing the feature dimension and reducing the overfitting problem of the model in high-dimensional data;(2)The optimized LightGBM is introduced to calculate the information gain for the eigenvalues.This module integrates a quantum genetic-based hyperparameter optimization algorithm to automatically adjust the LightGBM hyperparameters to achieve the purpose of improving model accuracy and model generalization ability;In addition,the regression coefficient of the feature value is obtained as the influence relationship on the dependent variable,and the information gain weight function is added to normalize the information gain of the feature,thereby extracting the influence degree of the influencing factors.After experimental verification,the proposed model improves the accuracy of influencing factor analysis with a small time cost.Finally,with the model proposed in this paper as a reference,a prototype system for analyzing the influencing factors of non-oil product sales in gas stations is designed and implemented,and the detailed design and test results of the system are expounded.