Green Energy Consumption Path Selection and Optimization Algorithms in the Era of Low Carbon and Environmental Protection Digital Trade

Abstract

In order to better study the chosen path of the consumption model of public green energy and more accurately predict consumers’ green energy consumer behavior, we take new energy vehicles as an example to explore the driving mechanism and internal mechanism of the public green energy consumption model from the perspective of motivation. We propose an ensemble learning model based on a stacking strategy. The model uses XGBoost, random forest and gradient lifting decision trees as primary learners to transform features, and uses logistic regression as a meta-learner to predict users’ consumer behavior. The experimental results show that this feature engineering method can significantly improve the accuracy rate in multiple model algorithms, and the prediction effect of the ensemble learning model is better than that of a single model, with the accuracy rate of 82.81%. In conclusion, the ensemble learning model based on a stacking strategy can effectively predict the public’s consumer behavior. This provides a theoretical basis and policy recommendations for promoting green energy products represented by new energy vehicles, thereby improving the practical path for proposing green energy consumption.

1. Introduction

Since entering the era of industrial civilization, humans have accelerated the plundering of natural resources in order to create material wealth. This has disrupted the balance of the Earth’s ecosystem, leading to increasingly prominent conflicts between humans and nature. In recent years, global warming, species extinction, frequent climate extremes and worsening desertification have posed a great threat to human survival and development. At the Leaders’ Climate Summit in April of this year, the Chinese President gave a significant speech titled “Jointly Building a Community of Life between Man and Nature” to address the unprecedented challenges facing global environmental governance. In his speech, he urged the world to adhere to the concept of harmonious coexistence between humans and nature and green development. Green development is a complicated and planned undertaking. Green consumption is an inevitable requirement for the construction of a community of human and natural life and is the key to fundamentally achieving green development.

The development of China’s economy has entered a new era. It will enter a stage of high-quality development, rather than a stage of rapid growth. Consumption is a major driving force for economic growth and plays a huge role in China’s economic growth. To develop in the direction of low-carbon sustainable economy of our country, it is necessary to develop in the direction of green, low-carbon and healthy consumption. Therefore, green consumption is the inevitable driving force for the transformation of our country’s economy to low-carbon sustainable development.

In recent years, along with the rapid development of our economy and the improvement of our national income level, the consumption level of residents has risen accordingly, which has pushed forward the upgrading of the consumption structure in our country. As a new consumption trend in recent years, green consumption conforms to the consumption concept of consumers pursuing quality and health and environmental protections and conforms to the development trend of consumption upgrading.

At present, although consumers’ awareness of green consumption has been significantly enhanced, their green consumption behavior is not enough to practice it [1].

Therefore, in order to better explore the consumption path choice of the public green energy consumption model and more accurately predict consumers’ green energy consumer behavior, this paper takes the new energy vehicle, a green energy product in the automotive industry, as an example, and proposes an ensemble learning model based on superposition strategy to predict users’ consumer behavior. Therefore, the practice path of green energy consumption is put forward.

Although some scholars define the essence of green consumption as a behavior with significant positive externalities and altruism, the classical theoretical models such as the theory of planning behavior and rational behavior theory are still widely used in the research field of green consumption. According to the theory of rational behavior, individuals will reasonably process all kinds of information they can obtain and take action only after comprehensively considering and weighing the meaning and results of a specific behavior. The behavioral intention of an individual is the most ideal variable to predict the actual behavior of an individual. It is jointly influenced by two major factors: attitude and subjective norms, while other factors need to indirectly influence the individual’s final behavior through attitude and subjective norms. In the analysis of individual environmental behavior, most scholars agree attitude can significantly affect behavior intention, and when the research attitude for specific environmentally friendly behavior (for example, green product purchase behavior, newspaper recycling behavior, etc.) rather than general environmental behavior, the attitude and behavior of the relationship between the intention will be stronger.

The cultivation path of consumer green consumption behavior is as follows:

• Shape values, integrating educational resources, guiding consumers to form a green consumption concept; give play to the education effect of good family traditions; adhere to the main position of school education; stimulate the vitality of community education; strengthen the positive education function of the media.
• Establish a system for the government to regulate and support green consumption behavior.
• Create a trend where enterprises lead the fashion of green consumption: enterprises carry out independent innovation in green technology and consolidate the production of green products; enterprises carry out green marketing, leading the trend of green consumption fashion.

China’s economic development has entered a new era, moving from a stage of high-speed growth to a stage of high-quality development. In order to adapt to the new stage of development, we must abandon the traditional extensive growth mode and pursue a “low-carbon, environmentally friendly and efficient” intensive growth mode. At the same time, we will optimize the industrial structure, promote the upgrading of the industrial structure, abandon the industrial structure which previously destroys the ecological environment and the export of low-end manufacturing and low value-added products, vigorously develop the medium- and high-end manufacturing industry, increase the export of high-value-added products, and gradually optimize the industrial structure. Consumption, as a major driving force for economic growth, plays a huge role in China’s economic growth. As China’s economy develops in a low-carbon and sustainable direction, consumption is bound to develop in a green, low-carbon and healthy direction. Therefore, green consumption is the inevitable driving force for China’s economy’s low-carbon and sustainable development.

In summary, in order to better explore the choice of the consumption path of the public green energy consumption model and predict consumer green energy consumption behavior more accurately, this article takes the green energy product in the automotive industry—new energy vehicles—as an example and discusses the driving mechanism and internal mechanism of the public green energy consumption model from the perspective of motivation. An integrated learning model based on stacking strategy is proposed. In the same data set, the feature engineering processing method of adding the original features extracted by the fixed length sliding window is used in the four classical model algorithms of supporting vector machine, logistic regression, XGBoost and random forest classification models. According to the experimental results, compared with the original features, the accuracy of the added features increased by about 5.75~14.92%. The above comparative experimental results show that the feature engineering method can improve the model effect in the paper—that is, it has certain effectiveness in the prediction of consumption behavior. At the same time, the advantages of multiple classifiers are fully learned and utilized to construct an integrated learning model for a stacking strategy to predict consumer behavior. The experiment proves that the integrated learning model has a better effect than a single model, with the highest accuracy rate of 82.81%, which fully verifies the effectiveness of the model in the prediction of consumer behavior.

The selection and optimization algorithms for green energy consumption paths in the era of low-carbon and environmentally friendly digital trade are of great importance, mainly reflected in the following aspect: reducing carbon emissions. The selection and optimization algorithms for green energy consumption paths can help minimize carbon emissions to the greatest extent possible. By optimizing the energy consumption path, renewable energy will be prioritized and the use of traditional high carbon energy will be minimized. This helps to reduce greenhouse gas emissions and address the challenges of climate change and global warming. Efficient utilization of resources, through algorithm optimization, can make more effective use of green energy resources. By accurately predicting energy demand and optimizing supply, energy allocation can be reasonably planned, renewable energy can be maximized and energy resource utilization efficiency can be improved. The improvement of economic benefits can be achieved through the selection and optimization algorithm of green energy consumption paths. By reasonably selecting energy consumption paths and optimizing energy allocation, energy consumption costs can be reduced, energy utilization efficiency can be improved and energy expenditure can be reduced. In summary, the importance of green energy consumption path selection and optimization algorithms in the era of low-carbon and environmentally friendly digital trade lies in achieving goals such as carbon emission reduction, effective resource utilization, economic efficiency improvement, digital trade development and environmental sustainability. Through scientific algorithm design and implementation, the optimization of green energy consumption can be achieved, providing support for sustainable development.

2. Literature Review

The concept of “green consumption” was first proposed by the International Consumer Union in 1963. Although the definition of the concept of green consumption behavior has not been unified in academic circles, the consensus of green consumption behavior has been reached in the following two aspects: firstly, green consumption behavior protects the environment; secondly, green consumption behavior minimizes the impact on consumers and environmental society and realizes the sustainable development of man and nature. There have been investigations on consumption path and consumption algorithms. Hojjat put forward the theory of rational behavior, which holds that individuals will reasonably process all kinds of information available to them and take actions only after comprehensively considering and weighing the significance and results of carrying out a certain behavior. The behavioral intention of an individual is the most ideal variable to predict the actual behavior of an individual, which is jointly affected by attitude and subjective norms, while other factors indirectly affect the final behavior of an individual through attitude and subjective norms [2]. In Stankovi’s analysis of individual environmental behavior, most scholars agree that attitude can significantly affect behavioral intention, and when the attitude in the investigation is aimed at specific environmentally friendly behaviors (such as green-product-buying behavior, newspaper-recycling behavior, etc.) rather than general environmental behavior, the correlation between attitude and behavioral intention will be stronger. Attitude is understood as the result of the evolution of related memory for something [3]. From the perspective of cognition, Agarwal defines low-carbon attitude as the integration of memory, cognition and emotion. However, ABC attitude theory proposed by psychologists holds that attitude is a relatively durable system with three dimensions including cognition, emotion and behavioral intention [4]. Maharum took the XGBoost algorithm as the feature transformation and input it into the logistic regression model. The experiment proved that the algorithm could well predict the purchasing behavior of users in e-commerce [5]. Kundu conducted experiments on the decision tree, support vector machine, random forest, fuzzy clustering, genetic algorithm and other technologies, and finally concluded that support vector machine is the best one [6]. Lai studied the changing relationship between consumer purchasing behavior and environmental factors, organizational factors, individual factors, interpersonal factors and other parameters, and designed a real-time evolutionary random forest classifier with unique feature engineering to predict consumer purchasing behavior [7]. Kirikkaleli et al. explored the relationship between globalization, GDP, GDP carbon intensity, patents and their impact on consumption-based carbon emissions (CCO2E). For the analysis, new econometric methods include nonlinear ARDL and Fourier ARDL, as well as robustness, dynamic OLS applications. The results of the co-integration test indicate a significant long-term relationship between CO₂ equivalent and globalization, economic growth, patent and GDP carbon intensity. Moreover, the empirical results show that only the positive impact of patents on environmental innovation has a significant negative impact on CCO2E, while the positive and negative impacts of GDP and GDP carbon intensity significantly increase CCO2E. However, only the negative shock of globalization would show an increase in CO₂ equivalent. Furthermore, the dynamic OLS results confirmed the robustness. Given this result, it is recommended that the Danish government be cautious in approving policies designed to promote economic growth, as this may have a negative impact on environmental sustainability [8]. In terms of algorithm modeling, Liobikien proposed that most of the existing experiments use a single model for prediction. Although there is also relevant literature on ensemble learning, they only use a classifier encapsulated with the idea of ensemble learning, such as catboost, random forest, XGBoost and other strong classifiers. Instead of making full use of the advantages of multiple classical classifiers, ensemble learning is used to combine these advantages for prediction [9]. Since the reform and opening up, China’s economy has developed rapidly and has made remarkable achievements. However, the traditional extensive development model of high investment, high consumption and high pollution has seriously damaged the ecological environment, resulting in various environmental problems such as massive energy and resource consumption, frequent sandstorms and haze and water pollution. The ecological environment has encountered unprecedented threats, and it is urgent to promote green development. Academics have different views on how to promote green development. However, many scholars believe that the development of green consumption is an important part of realizing green development. In the research in the field of green consumption, many scholars discuss the cultivation path of green consumption concept, and few scholars discuss how to cultivate consumers’ green consumption behavior. The cultivation of the concept of green consumption is important, but if it cannot be transformed into actual green consumption actions, the concept will become empty talk. The Fifth Plenary Session of the 18th CPC Central Committee put forward the five development concepts of “innovation, coordination, green, openness and sharing” in order to better handle the relationship between economic development and resource and environmental protection. In the background of vigorously promoting and practicing the concept of green development, cultivating a green consumption model is undoubtedly the key breakthrough. A green consumption model and advocating moderate conservation are not only in line with the connotation of the times of ecological civilization construction, but also can provide a strong driving force and lasting support for green development.

In order to better discuss the choice of consumer path of the public green energy consumption pattern and more accurately predict consumer green energy consumption behavior, taking new energy vehicles—a green energy product in the automotive industry—as an example, the driving mechanism and internal mechanism of the public green energy consumption pattern are discussed from the perspective of motivation. An integrated learning model based on the stacking strategy is proposed. The model uses the XGBoost, random forest and gradient lift decision trees as primary learning to transform features and uses logistic regression as meta-learning to predict user consumption behavior. The experimental results show that the feature engineering method can significantly improve the accuracy of multiple model algorithms. The integrated learning model has a better prediction effect than the single model, and the accuracy rate can reach 82.81%. The integrated learning model based on stacking strategy can effectively predict public consumption behavior. It is expected to provide theoretical basis and policy suggestions for the promotion of green energy products represented by new energy vehicles, so as to perfect the practice path of green energy consumption [10]. The conceptual model of variable measurement includes independent variables (internal motivation and external motivation), intermediary variables (green consumption attitude), dependent variables (purchase intention) and adjustment variables (consumer innovation) as shown in Figure 1.

3. Investigation Methods

3.1. Investigation Methods of Driving Mechanism of Green Consumption Pattern

3.1.1. Samples and Data

New energy vehicles were selected as the object of green product purchase intention, and the original questionnaire was revised based on the preliminary survey. The formal survey was carried out, and the questionnaires were distributed to 6 urban areas (A, B, C, D, E and F) of a certain city. The targets were medium and grassroots employees of government departments and enterprises and public institutions with certain sources of income. The survey adopted two methods to collect questionnaires. The second is to use Questionnaire Star, e-mail and other online tools to obtain data. In the whole investigation process, a total of 380 questionnaires were sent out and 356 questionnaires were actually collected, with a recovery rate of 93.7%. After eliminating the questionnaires with abnormal data, 323 effective questionnaires were obtained, with an effective rate of 85%.

3.1.2. Measurement of Variables

The conceptual model includes independent variables (internal motivation and external motivation), intermediary variables (green consumption attitude), dependent variables (purchase intention) and moderating variables (consumer innovation) [11], as shown in Figure 1.

A Likert 5-level scale was used to measure the questionnaire items. The 1–5 scale indicated how respondents felt from disagree to agree, where 1 represented “completely disagree” and 5 represented “completely agree”. The statistical analysis of the sample data shows that the proportion of males is higher than that of females, accounting for 59.8%. The age range is mainly from 31 to 50 years old. They are highly educated: all of them are educated to college degree or above, and their monthly disposable income is more than 4000 yuan. Firstly, in order to ensure the robustness of the analysis results, SPSS25.0 statistical software was used to analyze the reliability and validity of the scale data, and then the hypothesis proposed in this investigation was tested by a structural equation model. The principal component method was applied to conduct confirmatory factor analysis on variables involved in the model. The measured factor loads of each index and Cronbach’s α coefficient of each variable scale are shown in

Table 1. Confirmatory factor analysis and reliability test results.

Measuring Variables	Index of Measurement	Load of Factor	Value of Cronbach’s α
Internal motivation	Because like and buy	0.718	0.823
	Buying process is enjoyable	0.862
	Interested in new energy vehicles	0.641
External motivation	Some financial gains can be made	0.751	0.741
	Gain recognition and praise from others	0.737
	Can improve personal reputation	0.529
Consumer innovation	I like to try different things	0.907	0.915
	I can accept new things	0.813
	I often put forward different opinions to share with people around	0.952
	I know about the latest models of cars earlier	0.842
	I was an early buyer of a new model in my circle of friends	0.882
	I would be very interested in buying a new model car if it comes out	0.835

and

Table 2. Confirmatory factor analysis and reliability test results.

Measuring Variables	Index of Measurement	Load of Factor	Value of Cronbach’s α
Green consumption attitude	A wise choice	0.708	0.634
	Beneficial to everyone	0.692
	Promote new energy vehicles	0.550
Willingness to buy	Willing to collect and learn more information about new energy vehicles	0.853	0.872
	Willing to recommend relatives and friends to buy	0.869
	Willing to introduce and recommend to family members	0.719
	If necessary, willing to buy new energy vehicles	0.756

. It can be seen from Table 1 and Table 2 that the factor loading of all the measured items is more than 0.5, indicating a good fit of the model. The green consumption attitude is 0.6 < α< 0.7, and the reliability is acceptable; Cronbach’s α values of internal motivation, external motivation, consumer innovation and purchase intention were 0.823, 0.741, 0.915 and 0.872, respectively, all exceeding 0.7, indicating that the scale had high reliability and could reflect the situation of sample objects stably through internal consistency.

3.1.3. Structural Equation Model

With LISREL software, structural equation analysis was conducted on the relationship between self-determined motivation, green consumption attitude and purchase intention, and the maximum likelihood (ML) method was used to estimate the path value of the model so as to obtain the specific model fitting index shown in

Table 3. Goodness of fit index N = 323.

Fitting Index	χ2/df	GFI	AGFI	CFI	IFI	NFI	SRMR	RMSEA
Fitting Standard	<5	>0.90	>0.90	>0.90	>0.90	>0.90	<0.08	<0.1
Model	3.23	0.87	0.89	0.97	0.93	0.91	0.065	0.088
Fitting effect	ideal	Close	Close	ideal	ideal	ideal	ideal	ideal

[12]. It can be seen from Table 3 that the value of χ²/df (the ratio of chi-square value to the degree of freedom) is 3.23, which is less than the required standard critical value of 5. The values of GFI (goodness of fit index) and AGFI (modified goodness of fit index) are 0.87 and 0.89, respectively, which are not up to the recommended standard but are still close to the ideal value of 0.9. CFI (comparative fitting index) = 0.97 > 0.9, IFI (value-added fitting index) = 0.93 > 0.9, NFI (specification fitting index) = 0.91 > 0.9, SRMR (mean residual square root) = 0.065 < 0.08, RMSEA (difference square root) = 0.088 < 0.1 and other adaptation indexes meet the standards. On the whole, the degree of fitting between the investigation model and the empirical data is good, indicating that the model is acceptable and there is a good discriminant validity among all variables. Therefore, the standardized regression coefficient and t-value (the result of a t-test) in the model will be used in the following to analyze the degree and significance of the correlation between variables.

From

Table 4. Path coefficient and hypothesis testing of the new energy vehicle purchase intention model.

Suppose the Regression Path	Normalized Regression Coefficient	t-Value	Test of Hypothesis	Evaluation of Statistical Significance
Internal motivation—willingness to buy	0.22	2.21	support	significant
External motivation—willingness to buy	0.48	3.39	support	significant
Internal motivation—green consumption attitude	0.26	2.65	support	significant
External motivation—green consumption attitude	0.39	3.17	support	significant
Green consumption attitude—purchase intention	0.24	2.18	support	significant

, it can be seen that the positive effects of internal and external motivation on the willingness to purchase new energy vehicles have been verified, with corresponding t-values of 2.21 and 3.39, both greater than 1.96 at the 0.05 significance level. Therefore, hypothesis H1a and H1b have been verified [13]. Secondly, the significant influence of internal motivation and external motivation on green consumption attitude has also been confirmed, with corresponding t-values of 2.65 and 3.17, respectively, higher than the t-value of 1.96 at the significance level of 0.05. Therefore, H2a and H2b are assumed to be supported. This indicates that the enhancement of internal and external motivation is conducive to the positive green consumption attitude. Finally, the green consumption attitude has a significant impact on the purchase intention of new energy vehicles. The t-value is 2.18, which is greater than 1.96 at the significance level of 0.05, and H3 is verified.

3.1.4. Test of the Moderating Effect of Consumer Innovation

In this investigation, SPSS25.0 software is used to test the regulatory effect of consumer innovation, and the regulatory effect analysis can be achieved through three steps [14]. Therefore, in this investigation, internal motivation and external motivation were taken as independent variables, consumer innovation as a moderating variable, and green consumption attitude and purchase intention as dependent variables. SPSS25.0 statistical software (IBM, Armonk, NY, USA) was used to conduct hierarchical regression analysis on the two kinds of moderating effects. Specific test results are shown in

Table 5. Test results of consumer innovation regulation effect (willingness to buy).

Equation	Independent Variable	Coefficient of Normalization	t-Value	Sig	${\mathbf{R}}^2$	F Value
1	Internal motivation	0.247	2.274	0.000	0.149	9.401
	Consumer innovation	0.185	2.039	0.040		0.000
2	Internal motivation	0.234	3.442	0.000	0.158	11.756
	Consumer innovation	0.179	2.219	0.050		0.000
	Internal motivation × consumer innovation	0.131	2.067	0.258		0.000
3	External motivation	0.402	5.236	0.000	0.476	29.349
	Consumer innovation	0.247	2.326	0.002		0.000
4	External motivation	0.414	4.875	0.000	0.493	7.678
	Consumer innovation	0.256	2.373	0.002		0.000
	External motivation × consumer innovation	0.261	2.553	0.020		0.000

and

Table 6. Test results of consumer innovation regulation effect (attitude to green consumption).

Equation	Independent Variable	Coefficient of Normalization	t-Value	Sig	${\mathbf{R}}^2$	F Value
5	Internal motivation	0.270	4.197	0.000	0.149	9.401
	Consumer innovation	0.149	2.876	0.000		0.000
6	Internal motivation	0.263	5.546	0.000	0.158	11.756
	Consumer innovation	0.164	2.914	0.000		0.000
	Internal motivation × consumer innovative green consumption attitude	0.238	2.627	0.008		0.000
7	External motivation	0.336	3.974	0.000	0.476	29.349
	Consumer innovation	0.197	2.739	0.070		0.000
8	External motivation	0.354	3.162	0.000	0.493	7.678
	Consumer innovation	0.219	2.879	0.030		0.000
	External motivation × consumer innovative green consumption attitude	0.151	2.668	0.212		0.000

.

According to Table 5 and Table 6, the standardized regression coefficient of the product term of internal motivation and consumer innovation in Equation (2) is 0.131, which is significantly different from 0. We compared Equations (1) and (2) and found that ΔR² is 0.009, with a percentage change of 6.04%. It can be seen that consumer innovation plays a significant moderating role between the internal motivation and purchase intention of new energy vehicles. In the influence path of consumer innovation on external motivation and purchase intention of new energy vehicles, the t-values in Equation (3) are all greater than 1.96 at the 0.05 significance level, and the t-values in Equation (4) are 4.875, 2.373 and 2.553, respectively, which are all greater than 1.96 at the 0.05 significance level, passing the t-test. We compared Equations (3) and (4) and found that after adding the product term, the standardized regression coefficient is 0.261, and there is a significant difference between the regression coefficient of the product term and 0. ΔR² was 0.017, and the percentage change was 3.6%. It can be seen that consumer innovation has a significant moderating effect between external motivation and purchase intention of new energy vehicles. In the influence path of consumer innovation on internal motivation and green consumption attitude, the t-values in Equation (5) are 4.197 and 2.876, which are both greater than 1.96 at the 0.05 significance level. The standardized regression coefficient of the product term of consumer innovation and internal motivation in Equation (6) is 0.238, which is significantly different from 0. Comparing Equations (5) and (6), ∆R² is 0.019 and the percentage change is 11.45%. Therefore, consumer innovation plays a significant moderating role between internal motivation and green consumption attitude. In the influence path of consumer innovation on external motivation and green consumption attitude, the t-values before adding the product term are 3.974 and 2.739, and after introducing the product term, the t-values are 3.162, 2.879 and 2.668. All t-values are greater than 1.96 at the significance level of 0.05, and the standardized regression coefficient of the product term is 0.151. It is significantly different from 0, ∆R² is 0.018, and its change percentage is 18.18%, indicating that consumer innovation has a significant regulating effect between external motivation and green consumption attitude. Comparing Equations (7) and (8), ∆R² is 0.017, and the percentage change is 3.57%. Therefore, consumer innovation has a certain moderating effect on external motivation and green consumption attitude, but it is not obvious.

This investigation reveals that the public’s green consumption attitude and intention to purchase new energy vehicles are significantly influenced by both internal and external motivation. External motivation has a greater impact on green consumption attitudes and intention to purchase new energy vehicles than internal motivation [15]. The results of hypothesis testing are shown in

Table 7. Interpretive model corresponding to hypothesis.

Assumptions	Explanation	Assumptions	Explanation
H1a	Internal motivation positively influences the public’s green purchase intention	H4a	Consumer innovation plays a moderating role between internal motivation and green purchase intention
H1b	External motivation positively influences the public’s green purchase intention	H4b	Consumer innovation plays a moderating role between external motivation and green purchase intention
H2a	Internal motivation positively influences the public’s attitude towards green consumption	H4c	Consumer innovation plays a moderating role between internal motivation and green consumption attitude
H2b	External motivation positively influences the public’s attitude towards green consumption	H4d	Consumer innovation plays a moderating role between external motivation and green consumption attitude
H3	Internal motivation positively influences the public’s green purchase intention

and

Table 8. Investigation of driving mechanism of green consumption pattern.

Assumptions	Conclusion	Assumptions	Conclusion
H1a	support	H4a	regulate significantly
H1b	support	H4b	regulate significantly
H2a	support	H4c	regulate significantly
H2b	support	H4d	regulate significantly
H3	support

.

3.2. Construction of the Integrated Learning Model Based on Stacking Strategy

Ensemble learning combines multiple learners to complete learning tasks. Firstly, multiple individual learners are trained, and then some integration strategy is adopted to combine these individual learners. Multi-source data sets and multiple machine learning methods were used to construct a single detection model, and a logistic-based integrated learning method was designed to further improve the accuracy and robustness of the detection method for unknown variants of malware [16,17], as shown in Figure 2.

3.3. Evaluation Criteria

The experimental environment is the Windows10 operating system, with 16 G memory, 3.6 GHz 8-core processor, and the experimental software is Python3.7. This article uses four months of user consumption records provided by a 4S new energy vehicle service provider. One user purchasing one service represents one record. The invalid consumption data without user ID and noise data such as gifts, in-store and claims with unit price less than or equal to 0 are removed, and the remaining valid data total 2,059,718. According to the consumption duration of the first 30 days, whether the customer will finally order a car is predicted. If the customer has ordered a car in the last 7 days, it will be a positive sample; if there is no consumption, it will be a negative sample. Seven days will be taken as the step size and more sample data will be extracted by sliding. A total of 284,071 samples were extracted, including 15,018 positive and negative samples and 269,053 negative samples.

Considering the imbalance between the positive and negative samples of the experimental data, accuracy rate (PA) was used to evaluate the effectiveness of feature engineering in order to improve the reliability of the experiment. In the algorithm modeling effect verification experiment, the comparison of AUC value and ROC curve was added to make a more intuitive comparison of model effects [18]. The predicted values were classified and summarized to establish the confusion matrix as shown in

Table 9. Confusion matrix.

		Predicted Value
Confusion Matrix		Positive	Negative
True value	Positive	TP	FN
	Negative	FP	TN

. TP represents the number of users correctly predicted in the model that will respond to consumption; FP represents the number of users wrongly predicted in the model that will respond to consumption; TN represents the number of users wrongly predicted in the model that will not respond to consumption; FN represents the number of users wrongly predicted in the model that will not respond to consumption.

Among them, the calculation of accuracy, special effect and accuracy rate is shown in the following Formulas (1)–(3):

$$P=\frac{TP}{TP+FP}$$

(1)

$$S=TN/(TN+FN)$$

(2)

$$PA=\sqrt{\frac{TP}{TP+FP}\times \frac{TN}{TN+FN}}=\sqrt{P\times S}$$

(3)

The exactness rate P addresses the extent of the right forecast among all examples anticipated as sure models and measures the classifier’s capacity to distinguish positive models.

The classifier’s capacity to identify negative cases is measured by the specific degree S, which indicates the proportion of correct predictions among all samples predicted as negative cases.

The accuracy of the evaluation index PA is influenced by the accuracy of both responsive and non-responsive consumer users. This can avoid the impact of the deviation in the sample size between responsive and non-responsive consumer users on the evaluation indicators [19]. The AUC value is the area under the ROC curve. The larger the AUC value is, the larger the area under the ROC curve is, and the better the model effect is.

4. Result Analysis

4.1. Feature Engineering Validity Verification Experiment

In order to verify the effectiveness of the feature engineering method in the prediction of drug consumption behavior, the experimental steps are as follows.

In order to eliminate accidental problems, the ten-fold cross-validation method was adopted to conduct 10 experiments, and the results were averaged, as shown in Figure 3.

In the same data set, the feature engineering processing method is added to the original features extracted from the fixed-length sliding window. Ten-fold cross validation is used for comparison experiments in four classical model algorithms: the support vector machine, logistic regression, XGBoost and random forest classification models. The experimental results show that compared with the original features, the accuracy is improved by 5.75~14.92% after the feature engineering method is added.

4.2. Comparative Experiment of Integrated Learning Model

In order to verify that the ensemble learning model based on stacking strategy proposed in this paper is better than the single machine learning model, we choose the ensemble learning model to compare with the single model it contains. We compared the ensemble learning model with the XGBoost, random forest, gradient lifting decision trees and logistic regression [20]. The experimental results are shown in Figure 4 and Figure 5 and

Table 10. Comparison of accuracy between the model and a single model.

Model	Accuracy Rate%
XGBoost	69.65
Random forest	78.01
Gradient lifting decision tree	76.83
Logistic regression	72.55
Model in this article	82.81

.

The ROC curve comparison in Figure 4 shows that the overall performance of the integrated learning model based on the stacking strategy is better than the other four individual models in the prediction of consumption behavior, and the AUC value is the maximum, reaching 72.74%. From Table 10 and Figure 5, it can be seen that the integrated learning model based on the stacking strategy also has the highest accuracy, reaching 82.81%, which verifies the effectiveness of the proposed integrated learning model for the prediction of consumption behavior. In the same data set, the feature engineering processing method of adding the original features extracted by the fixed length sliding window is used in the four classical model algorithms of supporting vector machine, logistic regression, XGBoost and random forest classification models. According to the experimental results, compared with the original features, the accuracy of the added features is increased by about 5.75~14.92%. The above comparative experimental results show that the feature engineering method can improve the model effect in the paper—that is, it has certain effectiveness in the prediction of consumption behavior.

Based on this, the cultivation path of consumer green consumption behavior is as follows:

• Shape values, integrating educational resources, guiding consumers to form a green consumption concept; giving play to the education effect of good family traditions; adhere to the main position of school education, stimulating the vitality of community education; we will strengthen the positive education function of the media.
• Establish a system for the government to regulate and support green consumption behavior.
• Create a trend, with enterprises leading the fashion of green consumption; enterprises carry out independent innovation in green technology, and consolidate the production of green products; enterprises carry out green marketing, leading the trend of green consumption fashion.

5. Conclusions

The concept of “green consumption” was first proposed by the International Consumer Union in 1963. Although the definition of the concept of green consumption behavior has not been unified in the academic circles, the consensus of green consumption behavior has been reached in the following two aspects: firstly, green consumption behavior protects the environment; secondly, green consumption behavior minimizes the impact on consumers and environmental society and realizes the sustainable development of man and nature. In summary, in order to better explore the choice of the consumption path of the public green energy consumption model and predict consumers’ green energy consumption behavior more accurately, this article takes the green energy product in the automotive industry—new energy vehicles—as an example and discusses the driving mechanism and internal mechanism of the public green energy consumption model from the perspective of motivation. An integrated learning model based on a stacking strategy is proposed. In the same data set, the feature engineering processing method of adding the original features extracted by the fixed length sliding window is used in the four classical model algorithms of supporting vector machine, logistic regression, XGBoost and random forest classification models. According to the experimental results, compared with the original features, the accuracy of the added features is increased by about 5.75~14.92%. The above comparative experimental results show that the feature engineering method can improve the model effect in the paper—that is, it has certain effectiveness in the prediction of consumption behavior. At the same time, the advantages of multiple classifiers are fully learned and utilized to construct an integrated learning model for a stacking strategy to predict consumer behavior. The experiment proves that the integrated learning model has a better effect than a single model, with the highest accuracy rate of 82.81%, which fully verifies the effectiveness of the model in the prediction of consumer behavior.

Based on this, the cultivation path of consumer green consumption behavior is as follows:

• Shape values, integrating educational resources, guiding consumers to form a green consumption concept; giving play to the education effect of good family traditions; adhering to the main position of school education, stimulating the vitality of community education; we will strengthen the positive education function of the media.
• Establish a system for the government to regulate and support green consumption behavior.
• Create a trend; enterprises lead the fashion of green consumption; enterprises carry out independent innovation in green technology, and consolidate the production of green products; enterprises carry out green marketing, leading the trend of green consumption fashion.

Based on the driving mechanism model of the public purchase intention of new energy vehicles and the current industrial development environment, we propose the following suggestions in terms of the promotion and application path of new energy vehicles.

First, emphasize the economic benefits and social value of new energy vehicles, and develop positive incentive strategies based on external interests. On the one hand, economy is an important driving force for the public to buy new energy vehicles. As the national and local double subsidy policy gradually tightened to reduce the impact of the demand for new energy vehicles, consider “replace subsidies with awards” diversified policy support; explore ways such as support, ETC, preferential charging fees, exclusive parking spaces; improve the used car replacement residual value management measures and strengthen the new energy vehicles to the practical interests of the public. At the same time, we will support enterprises to cooperate with third-party service agencies to provide more value-added services such as rescue and insurance for new energy vehicle users. On the other hand, focus on the experience brought by social value to new energy vehicle users. Pay attention to the publicity of environmental attributes and symbolic significance of new energy vehicles, build an effective information exchange platform through various ways, encourage users of new energy vehicle to actively participate in interaction, express opinions and give timely affirmation and praise to valuable suggestions.

Second, weaken the public’s negative cognition of new energy vehicles, and pay attention to the cultivation of positive attitudes to the public. Although new energy vehicles have shown certain cost advantages in operation and maintenance, there are still many uncertainties in product maturity, technical performance and charging convenience. Therefore, first of all, the government can increase the financial support for R & D investment, encourage enterprises to break through the core and key technology difficulties, improve the competitiveness of product quality and promote the sustainable and healthy development of the new energy vehicle industry. Secondly, improve the supervision system, further tighten the access conditions, establish the product safety performance evaluation system and realize the whole life cycle monitoring of products. Finally, improve the construction of supporting infrastructure, improve the convenience and satisfaction of public use links, constantly optimize the number and location of public charging facilities and actively promote the construction of residential charging piles to solve the “last kilometer” problem of the promotion of new energy vehicles.

Limitations of the study:

• Data availability and quality: The data relied on by this study may be incomplete or inaccurate. Reliable green energy consumption data may be difficult to obtain and may vary among different regions and countries. This may affect the accuracy and reliability of the algorithm.
• Model assumptions: The algorithms and models used in the study may be based on some assumptions or simplifications for calculation and optimization. These assumptions may not fully reflect the actual situation, leading to limitations in practical application.
• Feasibility and implementation issues: Further research and exploration are needed to determine whether the algorithm and path selection methods proposed in this study are feasible in practice, as well as the implementation issues in actual environments. The limitations of the algorithm in terms of feasibility, adaptability and operability may require further verification and evaluation.

In summary, although this study proposes a green energy consumption path selection and optimization algorithm, its limitations need to be noted. In practical applications and further research, it is necessary to comprehensively consider issues such as data quality, multiple factors, model assumptions, algorithm complexity and feasibility, and conduct sufficient evaluation and validation.