Exploring the Role of Robots and Artificial Intelligence in Advancing Renewable Energy Consumption

Abstract

This study investigates the relationship between artificial intelligence (AI), industrial robots, and renewable energy consumption, driven by the rapid technological advancements and widespread adoption of AI tools in various industries. This research aims to evaluate the environmental implications of these technologies, specifically their impact on renewable energy usage. Employing a comprehensive analytical framework, this study utilizes advanced methodologies, including regularization factors, to accurately estimate the effects of these variables. Through a thorough data analysis, the research quantifies how AI and industrial robots influence the shift towards renewable energy sources. The findings reveal that investments in AI significantly enhance renewable energy consumption, as demonstrated by both conventional estimation techniques and those that integrate regularization factors. Conversely, the use of industrial robots is found to have a detrimental effect on renewable energy consumption. These results have important implications for policymakers, industry leaders, and sustainability researchers. This study encourages policymakers and investors to prioritize funding for AI solutions that promote renewable energy adoption, while it advises industry managers to strategically modify their use of industrial robots to reduce their environmental impact. Ultimately, this research lays a critical foundation for future inquiries and policy initiatives aimed at aligning technological advancements with sustainable energy practices.

1. Introduction

The global energy landscape is undergoing a transformative shift towards renewable sources, driven by the urgent need to combat climate change and reduce our dependency on fossil fuels [1]. In this evolving paradigm, robots and artificial intelligence (AI) are emerging as pivotal technologies that could revolutionize how we generate, distribute, and consume renewable energy [2]. From optimizing wind farm operations to enhancing solar panel efficiency and managing smart grids, AI and robots are poised to address some of the most pressing challenges in the renewable energy sector [3]. This article explores the intersection of robots, AI, and renewable energy, examining how these advanced technologies can accelerate the transition to a sustainable energy future. By harnessing the power of intelligent systems, we can not only improve the efficiency and reliability of renewable energy sources but also pave the way for innovative solutions that drive the widespread adoption and integration of green energy into our daily lives.

In Dincer’s view [4], “energy is the convertible currency of technology”, and in assessing the level of economic growth of a country, the relationship between energy production and consumption is crucial, as emphasized by Pramanik [5]. Energy consumption is closely linked to economic activities and production, and the availability and efficiency of energy resource utilization can significantly influence a nation’s economic performance. An increase in energy production can support economic expansion by creating jobs, stimulating investments, and supporting infrastructure development. On the other hand, efficient and sustainable energy consumption can provide a solid foundation for long-term economic growth by reducing production costs, improving competitiveness, and protecting the environment. Thus, understanding and appropriately managing the relationship between energy production and consumption are crucial aspects in promoting sustainable economic development and ensuring energy security in an increasingly interconnected world dependent on energy resources.

In the context of limited resources and increased CO₂ emissions, humanity is seeking sustainable solutions to address these challenges, and renewable energy represents the only practical and accessible alternative to conventional energy sources. This includes various forms, such as hydropower, geothermal energy, tidal energy, solar energy, and wind energy [6]. Interest in renewable energy emerged in the 1970s during the energy crisis when the fear of depleting conventional fossil fuels led to intense programs for the development of renewable sources and energy conservation measures [7]. In addition to these factors, population growth and energy demand per capita of various types add to the pressure [8]. According to a forecast by the International Institute for Applied Systems Analysis and the World Energy Council, it is estimated that the world’s population will reach 10 billion by the mid-21st century. Additionally, it is estimated that the global demand for energy services will significantly increase by 2050, while primary energy demand is anticipated to increase 1.5 to 3 times over [9]. Given that electricity is a crucial factor for global progress, numerous researchers have dedicated themselves to generating electricity from renewable sources to counteract the electricity deficit in households and industrial areas [10].

As a result of the need to increase its flexibility and reduce its costs and environmental impact (by optimizing resource consumption and reducing CO₂ emissions), the global energy industry is undergoing substantial changes in terms of energy generation, distribution, storage, and trading methods [11]. In these circumstances, AI has a significant impact on the development and implementation of clean energy sources. AI is primarily focused on developing intelligent machines and software for specific problems in various fields, one of which is the energy sector [12]. AI also plays a substantial role in advancing technologies in the renewable energy domain. Many aspects of renewable energy, such as its design, development, assessment, operation, distribution, and regulation, significantly depend on the use of AI [5]. Through advanced machine learning algorithms and data analysis, AI optimizes the efficiency and reliability of renewable energy systems, contributing to accelerating the transition to more sustainable and environmentally friendly energy sources.

According to Kow et al. [13], in numerous countries, AI has been integrated into performing various tasks, including the control, forecasting, and efficient operations of renewable energy systems. This integration of AI materializes as tangible benefits, such as facilitating efficient inverter control in photovoltaic systems [14] and optimizing the ability to track power points [15], as well as carrying out efficient forecasting for solar, wind, hydro, and geothermal energy [16,17,18].

Lytras and Serban argue that improving the design of energy infrastructure and the implementation and production of renewable energy is necessary and achievable only through the introduction of AI at all levels in energy networks, leading to their further development [19].

The use of renewable energy resources is essential to meet the increased demand for energy sustainably without compromising natural resources and exacerbating climate change. It is important for governments, companies, and individuals to invest in the development and utilization of these clean energy sources to ensure a safer and more sustainable future for future generations. In this context, robots and AI become essential tools in the development and implementation of innovative solutions promoting the sustainable use of renewable energy resources; providing support in the design and management of renewable energy, which is vital for combating climate change; ensuring environmental sustainability; and promoting a more resilient energy economy.

This study aims to investigate the impact of artificial intelligence and robots on renewable energy consumption in European countries, focusing on the nature of this relationship (positive or negative). The novelty of this research lies in its pioneering application of elastic net (ENET) regression techniques to evaluating the influence of AI on renewable energy consumption. This advanced analytical framework provides a more refined understanding of the interaction between AI and renewable energy utilization, surpassing the constraints of traditional regression methods. Additionally, this study’s originality is highlighted by its examination of a previously unexplored sample in relation to the research objectives. By addressing this gap in the existing literature, this study not only offers valuable insights but also establishes a benchmark for comparative analysis. This approach not only enhances our understanding of the impact of AI on renewable energy within the region studied but also offers broader implications for diverse geographical contexts.

In this context, the content of the paper is structured into five key sections. Section 2 provides a comprehensive review of the existing literature, with a particular focus on how robots and artificial intelligence influence the consumption of renewable energy. Section 3 focuses on describing the methodology adopted in this study, detailing the datasets used and the analysis methods implemented. Section 4 presents the research findings and offers an in-depth interpretation of the data, highlighting the main implications of these findings. Finally, Section 5 summarizes the key conclusions of this study, discusses their impact on public policy, and highlights the specific contributions of this research, while also providing recommendations for future research directions.

2. Literature Review

Interest in renewable energy in the research has emerged alongside an awareness of the gravity of the global issue regarding the limited nature of the material resources used in the energy emission process and the greenhouse gas emissions resulting from the use of traditional energy. Both decision-makers and researchers have focused on identifying sustainable alternatives to traditional energy, as well as the factors influencing these renewable sources. Thus, alongside renewable energy sources, potential catalysts in the use and consumption of this type of energy have been identified, with one of them being artificial intelligence. AI is a term referring to a collection of computerized systems performing activities typically carried out by humans [20].

According to Das et al. [21], the integration of AI into the energy sector is essential, as it makes significant contributions by enhancing the monitoring, exploitation, maintenance, and storage of renewable energy, as well as implementing real-time operating and control systems. In detail, the authors identified several major applications of AI to renewable energy integration: renewable energy generation, considering the variability in renewable sources and the volatility of their supply; ensuring network stability and reliability, including security operations; precise forecasting of demand and weather conditions; efficient energy demand management; the optimization of energy storage operations; the development and management of energy markets; and increasing the connectivity between network components and with microgrids.

It has also been found that the use of AI, through applied machine learning, optimization, and intelligent communications, has a positive impact on increasing renewable energy productivity, while also ensuring cost reductions and introducing innovations into smart grids [22,23].

Ahmad et al. [24] argue that the integration of energy supply and demand and renewables into the grid will be autonomously managed by advanced software that will improve decision-making and operational efficiency, with AI playing a key role in achieving this goal. Using AI to drive green technological advancements can enhance the deployment of renewable energy and boost energy efficiency. This shift can transform national energy frameworks from the traditional, high-pollution, and high-consumption models into cleaner, environmentally friendly alternatives, thereby increasing the demand for renewable energy products [25].

Sahota [26] has also identified a broad set of benefits of AI for renewable energy: extensive data exploration and project viability forecasting; carbon emission reductions; providing optimized performance; energy production forecasting; and energy storage. These advantages of AI in the renewable energy sector further accelerate its integration. The benefits of AI regarding energy forecasts have been extensively studied, with particular attention given to each type of renewable energy source, such as photovoltaics (PVs), wind, geothermal, biomass, hydropower, etc. [27,28,29,30,31]. For example, to predict the wind speed used for electricity generation, especially in wind turbines, fuzzy modeling is used, as well as artificial neural network (ANN) technologies [32]. To approximate solar radiation, even on terrain, Bosch et al. [33] proposed an artificial intelligence technique using an ANN. Some other algorithms for predicting solar radiation include both ANN and neuro-fuzzy inference systems [34]. Furthermore, Fan [35] asserts that technological innovations driven by artificial intelligence can expand the scope of searching for and presenting a country’s commercial information, enhance the alignment between the supply and demand for renewable energy products, and thus also contribute to reducing trade barriers.

In addition to these advantages, Alankrita and Srivastava [27] have also addressed the main limitations of using AI in renewable energy. One of the main challenges with machine learning systems is the high cost of the specialized equipment and the need for careful data processing, as data processing and cleaning can be operations with significant costs. Additionally, machine learning is highly susceptible to bias. In the case of machine learning, this refers to the possibility that algorithms may be distorted or produce inaccurate results due to biased input data or algorithmic settings, which can compromise the entire model. Therefore, careful design and implementation of machine learning systems are crucial, and with a proper model, many of the issues associated with renewable energy systems can be addressed.

Tomazzoli et al. [36] introduced using AI in a study to enable energy efficiency. The authors argue that this approach could be used to develop a new architectural framework for the system to enable unified energy efficiency in dispersed networks of electrical appliances. It is also used to develop operating rules and to determine the type of computer that provides the best practice in this regard. The automated implementation of such an energy efficiency system can have significant implications in the field of smart energy, providing energy managers with the ability to efficiently control and configure a large number of subdivisions.

The application of AI in the use of renewable energy has multiple effects on achieving targets within the sectors of the environmental, society, and the economy. This is due to AI’s ability to improve the operation and efficiency of RE sources and reduce the operating costs and produced energy, as well as efficiently minimizing their impact on the environment [37,38]. Furthermore, Ahmad et al. [24] found that energy industry players, including utilities, system operators, and independent power producers, need to focus more on AI technologies to achieve significant results and remain competitive.

Fan et al. [39] highlighted the wide range of functionalities covered by AI in renewable energy, from energy forecasting and anomaly detection in energy systems to more intricate applications, such as designing renewable energy systems and network stability. The importance of efficient AI and models with a deep understanding in terms of energy has also been emphasized by Strubell et al. [40], highlighting their critical role in renewable energy research. In their study, Song et al. reveal that analysis of its mechanisms of action shows that AI can mitigate the vulnerability of the renewable energy supply chain through technological innovation, optimization of the governance systems, and enhancements to the trade network status [41].

The introduction of robots into manufacturing environments and other applications results in a remarkable improvement in productivity and efficiency [42,43]. AI and robots are playing a crucial role in the technological revolution of the 21st century, transforming the way we interact with technology and resources and opening up new possibilities for innovation and sustainable development [44]. Given that robots have been implemented for a wide range of applications, including operations and manufacturing [45,46,47,48], we considered it essential to include the variable “ROBOTS” in our analysis to assess the impact of automation on renewable energy consumption. Previous studies indicate that automation, including the use of robots, can optimize industrial processes and reduce energy consumption [49,50]. Robots are also frequently used in the installation and maintenance of renewable energy equipment, influencing efficiency and costs [51]. Thus, the inclusion of this variable allow us to investigate whether an increase in the use of robots has a significant effect on the consumption of renewable energy, reflecting the importance of automation technologies in this field.

According to these studies, the analysis of the relationship between renewable energy, AI, and robots has focused on the applicability of and benefits brought about by using AI and robots in managing and forecasting renewable energy, as well as the advantages resulting from the integration and use of these technologies in the renewable energy sector. While existing studies have established the potential benefits of AI and robots for specific renewable energy applications, there is a lack of comprehensive research that integrates these technologies’ impact on renewable energy consumption across different regions. Additionally, most studies have employed traditional regression techniques, which may not capture the complex interactions between AI, robots, and renewable energy consumption. This study addresses these gaps by applying advanced regression techniques, ENET regression, to analyzing the relationship between AI, robots, and renewable energy consumption in European countries.

3. Materials and Methods

3.1. Data Description and Characteristics

To investigate the effect of AI and robots on renewable energy consumption, we utilized a series of selected and processed indicators from various sources for the period 2012–2022, focusing on European countries (

Table 1. Data used in this study.

Indicator Name	Acronym	Variable Type	Unit of Measurement	Source
Share of energy from renewable sources	REN	Dependent variable	Percentage	Eurostat
Annual installation of industrial robots	ROBOTS	Explanatory variable	1000 units	International Federation of Robotics
Number of companies in AI	COMPANIES	Explanatory variable	Number	Artificial Intelligence Index Report
Total investment in artificial intelligence	TOTAL_INV	Explanatory variable	Billions of US dollars	Artificial Intelligence Index Report

). This extensive dataset enabled us to thoroughly analyze trends and patterns over the specified timeframe, offering valuable insights into the dynamics of the variables under study across the European region. To achieve this, we selected the share of energy from renewable sources as the dependent variable, measured as the percentage of total energy consumption. The data for this variable were sourced from the Eurostat database [52]. As explanatory variables, we included the annual number of industrial robot installations (ROBOTS), as reported by the International Federation of Robotics [53]. Additionally, from the AI domain, we incorporated two key indicators: the number of companies operating in artificial intelligence (COMPANIES) and the total investment in AI (TOTAL_INV), expressed in billions of US dollars and adjusted for inflation, based on data from the Artificial Intelligence Index Report [54]. These variables were chosen to capture the influence of technological advancements on renewable energy adoption across Europe.

Descriptive statistics analysis presented in

Table 2. Descriptive statistics of the variables.

	REN	ROBOTS	COMPANIES	TOTAL_INV
Mean	16.19648	64,400.00	184.1000	4.358118
Median	15.27799	66,500.00	197.0000	2.584934
Maximum	19.29363	84,000.00	343.0000	12.50078
Minimum	13.74178	43,000.00	53.00000	0.526805
Std. Dev.	2.064102	14,923.51	98.01185	4.321249
Skewness	0.557412	−0.146591	0.138157	0.964644
Kurtosis	1.720136	1.599229	1.767289	2.475798
Jarque–Bera	1.200369	0.853381	0.664969	1.665391
Probability	0.548710	0.652665	0.717140	0.434875
Sum	161.9648	644,000.0	1841.000	43.58118
Sum Sq. Dev.	38.34465	2 × 10⁹	86,456.90	168.0588

is essential for understanding the data used in the model. It provides an overview of their distribution, detecting anomalies and highlighting the relationships between variables. It helps us identify relevant data and validate our hypotheses. By summarizing and exploring the data in depth, descriptive statistics analysis serves as a basis for decision-making and further statistical analyses.

The results of the Jarque–Bera normality test offer insights into the distribution characteristics of the data utilized [55]. Specifically, for the variable REN, the Jarque–Bera test statistic yields a value of 1.200369, with an associated probability of approximately 0.548710. This suggests insufficient evidence to reject the null hypothesis that the REN data adhere to a normal distribution. Similarly, concerning the variable ROBOTS, the Jarque–Bera test produces a statistic of 0.853381, with a probability of around 0.652665, indicating a lack of evidence to reject the null hypothesis of normality. Likewise, for the variables COMPANIES and TOTAL_INV, the Jarque–Bera test results similarly indicate a lack of evidence to reject the null hypothesis of a normal distribution given the relatively high associated probability values. In summary, according to the Jarque–Bera test, the data for all the variables appear to approximate normality, as there is insufficient evidence to consider them to deviate from a normal distribution.

3.2. The Elastic Net Regularization Regression Model

ENET is widely embraced as a remedy for overfitting, a scenario in which a model demonstrates a strong fit with the training data but struggles to generalize effectively to novel test data. The preservation of and a reduction in the magnitudes of the regressors in the elastic net model hinge upon the specific parameters selected for the regularization process.

Since its introduction, ENET has become widely used in statistics, machine learning, and various other fields due to its ability to handle high-dimensional data and guard against overfitting [56]. It adds a penalty term to the ordinary least squares (OLS) objective function, which encourages the coefficients of less important variables to be exactly zero. This leads to sparse models where only a subset of predictors are chosen, while the others are effectively ignored [57].

Although the OLS estimator is valued for its unbiasedness and other beneficial attributes, it can exhibit high variance under certain conditions. For instance, if the dataset contains more predictors, or if many predictors are highly correlated, the least squares estimates become highly sensitive to random errors and may exhibit substantial variance. Regularization techniques can be employed to mitigate this variance by introducing bias, thereby reducing the overall error, as seen in the elastic net regression model.

Elastic net, Lasso, and ridge regression are examples of penalized regression methods that function by diminishing the magnitudes of the model’s coefficients. Typically, this is achieved by adjusting the standard cost function of linear regression to include a penalty term [44].

$$J={\displaystyle \frac{1}{2m}}{\sum _{i=1}^m\left(y_i-{\beta }_0-\sum _{j=1}^p{x}_{ij}{\beta }_j\right)}^2+\lambda \left[{\displaystyle \frac{(1-\alpha )}{2}}\sum _{j=1}^p{\beta }_j^2+\alpha \sum _{j=1}^p\left|{\beta }_j\right|\right]$$

(1)

where

m is the number of samples;

p is the number of predictors;

y_i is the target variable for the ith sample;

xij is the value of the jth predictor for the ith sample;

β_j is the coefficient associated with the jth predictor;

β₀ is the intercept term.

The size of the penalty parameter λ determines the strength of the penalty’s influence. When λ is set to a high value, minimizing the cost function will result in shrinking the values of the coefficients β, potentially reducing some to zero. This leads to a simpler model with smaller (or zero) coefficients, making it less susceptible to overfitting. It is important to note that the penalty does not apply to the constant term β₀; the goal is solely to reduce the magnitudes of the regression coefficients.

3.3. The Methodological Framework

The stages of the methodology employed in this study are illustrated in Figure 1, which provides a visual overview of the analytical process. The first step involves assessing the stationarity of the data, a crucial aspect for time series analysis, conducted using the Augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) unit root tests. The null hypothesis of these tests is that the time series has a unit root. The p-value indicates the probability of observing in the data whether the null hypothesis is true. A smaller p-value (typically below a chosen significance level like 0.05) indicates stronger evidence against the null hypothesis [58]. This assessment is essential to verify that the data are suitable for subsequent analyses, as non-stationary data can lead to spurious results.

Following the stationarity tests, we examine the autocorrelation within the data through a correlation matrix analysis. This step is vital for understanding the relationships between the variables and identifying any potential lagged effects that may influence the outcomes of our models. To further ensure the integrity of the subsequent regression analyses, we rigorously check for multicollinearity, heteroscedasticity, and the distribution of the residuals. These diagnostic checks help confirm that the assumptions underlying the regression analysis are met, thereby increasing the reliability of our findings.

In estimating the initial model parameters, we utilize OLS regression, which serves as a foundational approach to understanding the relationships among the variables. Running OLS regression before ENET allows us to perform diagnostic checks on the data and assess the assumptions of linear regression. This can include checking for multicollinearity, heteroscedasticity, and examining the distribution of the residuals. OLS regression helps us select relevant predictors for inclusion in ENET. We use techniques like stepwise selection or feature importance ranking based on the OLS coefficients to identify important variables before applying regularization. The results obtained from the OLS regression set the stage for more complex estimation, which combines the strengths of both Lasso and ridge regression techniques. Within the ENET framework, the optimal model selection is achieved through a comprehensive evaluation of how the model coefficients evolve across various regularization parameters. This process is supported by graphical representations that facilitate the visualization of the coefficient paths.

To assess the predictive performance of the ENET model, we employ cross-validation techniques, with the errors graphically represented to evaluate the model’s reliability and robustness. This thorough evaluation ensures that our final analytical outcomes are both reliable and actionable. The summary path of ENET typically illustrates the evolution of the model coefficients as the regularization parameter varies. It provides insights into how the coefficients change relative to the strength of regularization, highlighting the trade-off between the model’s complexity and performance. This visual representation aids in selecting the optimal regularization parameter and understanding which variables contribute to the model’s predictive power the most. Next, we will examine how the coefficients evolve concerning the lambda penalty expressing the model’s complexity, and we present a graphical representation of the cross-validation errors, with the lambda path along the x-axis and the mean error measures of both the training and test sets along the y-axis. Finally, to identify causal relationships and their directions within the model, we apply the Granger causality test. This test allows us to explore whether one variable can predict another over time, providing further insights into the dynamics between the variables under investigation. Through these meticulous methodological steps, we aim to deliver a comprehensive and credible analysis of the factors influencing renewable energy adoption in the context of AI and robotics.

4. Results and Discussions

Before performing an ENET analysis, it is useful and necessary to apply several preliminary tests and analyses to ensure that the data and the model are properly prepared. We will first apply tests of the stationarity of the time series data, proposing the ADF and PP tests.

The results in

Table 3. Standard unit root tests.

	ADF		PP
	t-Statistic	p-Value	t-Statistic	p-Value
REN	−3.5703	0.0323	−3.5834	0.0317
ROBOTS	−3.2149	0.0533	−5.0831	0.0043
COMPANIES	−4.1142	0.0216	−6.4591	0.0013
TOTAL_INV	−3.4106	0.0512	−4.4464	0.0118

represent the outcomes of the ADF and PP tests for the stationarity of different time series variables. The t-statistic value indicates the strength of the evidence against the null hypothesis. The more negative the t-statistic, the stronger the evidence for rejecting the null hypothesis of a unit root, suggesting stationarity. Overall, based on these results, all of the variables appear to be stationary, with some variability in the level of statistical significance across variables.

It is also necessary to check the data for autocorrelation.

Table 4. Correlation matrix.

Probability	REN	ROBOTS	COMPANIES	TOTAL_INV
REN	1.000000
	-----
ROBOTS	0.758145	1.000000
	0.0110	-----
COMPANIES	0.882241	0.951875	1.000000
	0.0007	0.0000	-----
TOTAL_INV	0.869751	0.835128	0.916303	1.000000
	0.0011	0.0026	0.0002	-----

represents the results of the correlation matrix for the variables in the model. The results represent the covariance analysis in the form of correlation coefficients between different variables. Based on the results provided, there is no indication of autocorrelation between the variables.

According to the results for the OLS estimation provided in

Table 5. OLS estimation results.

Dependent Variable: REN
Method: Least Squares
Variable	Coefficient	Std. Error	t-Statistic	Prob.
ROBOTS	−0.000174	6.79 × 10−5	−2.564183	0.0504
COMPANIES	0.050697	0.015997	3.169180	0.0248
TOTAL_INV	2.286939	1.096942	2.084832	0.0915
C	19.33905	2.468732	7.833598	0.0005
R-squared	0.920185	Mean dependent var		16.19648
Adjusted R-squared	0.856333	S.D. dependent var		2.064102
S.E. of regression	0.782367	Akaike info criterion		2.653867
Sum squared resid	3.060489	Schwarz criterion		2.805159
Log likelihood	−8.269333	Hannan–Quinn criter.		2.487899
F-statistic	14.41116	Durbin–Watson stat		2.316374
Prob (F-statistic)	0.005940

, the coefficient for the variable ROBOTS is approximately −0.000174. This indicates that when holding the other variables constant, a one-unit increase in the ROBOTS variable is associated with a decrease of approximately 0.000174 units in the dependent variable REN. The coefficient for the variable COMPANIES is approximately 0.050697, suggesting that an increase of one unit in COMPANIES, when holding the other variables constant, leads to a rise of about 0.050697 units in REN. Similarly, for TOTAL_INV, the coefficient is approximately 2.286939, indicating that a one-unit increase in TOTAL_INV, with the other variables held constant, results in an increase of approximately 2.286939 units in REN. All the results are statistically significant at the 10% level.

The constant C in the table presenting the OLS regression results has a value of 19.33905 and is statistically significant, with a p-value of 0.0005, well below the common threshold of 0.05. This suggests that in the absence of an influence from the independent variables ROBOTS, COMPANIES, and TOTAL_INV, the expected value of the dependent variable REN would be approximately 19.34 units. This constant reflects the baseline level of REN and is significantly different from zero, indicating an essential component of the model in explaining the variation in REN.

The R-squared value (0.920185) indicates that approximately 92.02% of the variability in the dependent variable REN is explained by the independent variables included in the model. The F-statistic (14.41116) tests the overall significance of the regression model. A small p-value being associated with this statistic indicates that the model as a whole is statistically significant. Also, the Durbin–Watson statistic tests for autocorrelation in the residuals. A value of around 2 suggests no autocorrelation.

ENET offers a powerful framework for addressing many of the limitations of OLS regression, including multicollinearity and overfitting. It strikes a balance between bias and variance and provides improved predictive performance and interpretability in many practical scenarios. The results for the ENET estimation are presented in

Table 6. ENET estimation results.

Dependent variable: REN
Method: Elastic net regularization
Sample (adjusted): 2013 2022
Penalty type: Elastic net (alpha = 0.5)
Lambda at minimum error: 0.3376
Regressor transformation: Std dev (pop)
Cross-validation method: K-fold (number of folds = 10), rng = kn, seed = 1,823,593,784
Selection measure: Mean squared error
	(minimum)	(+1 SE)	(+2 SE)
Lambda	0.3376	1.363	1.977
Variable		Coefficients
C	13.95345	14.81947	15.38765
ROBOTS	0.000000	2.88 × 10−6	2.25 × 10−7
COMPANIES	0.007130	0.003378	0.002254
TOTAL_INV	0.111156	0.064610	0.042904
d.f.	3	4	4
L1 Norm	14.07174	14.88746	15.43281
R-squared	0.778264	0.585350	0.420285

.

We use ENET with a balance parameter (alpha) of 0.5, which indicates a combination of L1 (Lasso) and L2 (ridge) penalties. The value of the lambda regularization parameter that minimizes the mean squared error is 0.3376. Standard deviation (population) transformation was applied to the regressors. As the cross-validation method, we take K-fold cross-validation with 10 folds, using a specific random number generator (rng) and seed for reproducibility. The mean squared error was used to select the optimal lambda value.

The coefficients represent the estimated effects of each predictor variable on the dependent variable REN under the specified regularization. For ROBOTS, the coefficient suggests that there is minimal effect on REN when the number of robots increases by one unit, as the coefficient is close to zero. However, an increase in the regularization factor leads to a more robust model in which we identify a negative influence of the use of industrial robots on renewable energy consumption. This is primarily due to the increase in the overall energy demand, which, if not adequately supported by renewable sources, can lead to a greater reliance on non-renewable energy sources and the diversion of financial resources away from the development of renewable energy.

A one-unit increase in the variable COMPANIES is associated with an increase in REN by approximately 0.007130 units (minimum lambda). Growth in the number of AI companies can significantly boost renewable energy consumption by driving innovation and technological advancements in energy systems. As competition among these companies fosters the development of new and efficient technologies, such as optimized solar panels and wind turbines, it leads to reduced operational costs and enhanced efficiency. Additionally, AI firms contribute to improved energy management through advancements in smart grids and energy storage solutions, attract increased investment, and offer sophisticated data analysis and decision-making tools. This collective impact enhances the overall effectiveness and adoption of renewable energy, making it a more viable and competitive alternative to traditional energy sources.

An increase in TOTAL_INV by one unit is associated with an increase in REN by approximately 0.111156 units (minimum lambda). With lambda at the minimum error, the R-squared value is 0.778264, indicating that approximately 77.83% of the variability in the dependent variable REN is explained by the predictor variables under the specified regularization. Investment in AI can substantially boost renewable energy consumption by enhancing energy production efficiency through the predictive maintenance and optimization of renewable assets like solar panels and wind turbines. AI facilitates better energy management with smart grids and accurate demand forecasting, improves energy storage solutions through optimized battery management and grid integration, and supports informed policy decisions and investment strategies via advanced data analysis and simulations. Additionally, AI-driven automation reduces operational costs while fostering innovation in new technologies and integrating advancements from various fields, ultimately driving increased adoption and utilization of renewable energy sources.

The constant C has coefficients ranging from 13.95345 to 15.38765 across different levels of lambda, indicating that the baseline value of the dependent variable REN (when all the independent variables are standardized and penalization is applied) lies within this range. The variation in C across the lambda values reflects the impact of regularization on the model, where higher penalization leads to a slightly higher intercept. Despite regularization, the intercept remains a substantial contributor, signifying a consistent baseline level of REN independent of the specific influence of ROBOTS, COMPANIES, and TOTAL_INV under the constraints of the ENET model.

Table 7. Summary path.

	Lambda	d.f.	L1Norm	R-Squared	AIC
1	3.455171	0	16.19648	0.000000	0.840630
2	3.148223	3	16.06601	0.079768	1.373575
…
25	0.370487	3	14.11308	0.774645	0.789440
26	0.337574	3	14.07174	0.778264	0.786398
27	0.307585	3	14.03272	0.781359	0.783796
…
99	0.000379	4	19.17368	0.900273	0.883833
100	0.000346	4	19.18435	0.900251	0.883852

shows the lambda path in the left column, the model’s degrees of freedom in the second column, the L1 norm of the coefficients in the next column, and the R-squared of the model.

In Figure 2, we examined how the coefficients evolve concerning the lambda penalty. As anticipated, with higher penalization, the model’s complexity diminishes, causing the coefficients to gradually approach zero. According to cross-validation, the model at +1 SE (lambda = 1.363) is selected, coinciding with the point where most of the coefficients have been eliminated from the model.

Subsequently, we present a graphical representation of the cross-validation errors in Figure 3. As anticipated, the training error consistently remains lower than the test error, indicating the model’s better performance on the data it was trained on compared to unseen data. This observation aligns with the typical behavior expected during model evaluation and validation processes.

Our results have shown that investments in AI, characterized by the variables TOTAL_INV and COMPANIES, in the field of renewable energy have positive implications, which is consistent with the findings of other researchers, such as [22,23], who demonstrated that the use of AI has a positive impact on increasing the productivity of renewable energy, facilitating optimization, and ensuring cost reductions. Abdelshafy et al. [59] investigated the optimization of renewable energy storage based on AI, with their results indicating that the genetic algorithm (GA), which is an AI technique, optimizes the integration of photovoltaic, wind, and pumped storage energy in terms of the economic and environmental performance of the system. Additionally, a study conducted by García-Triviño et al. [60] showed that implementing the particle swarm optimization (PSO) algorithm in renewable energy management systems provides both a 29.36% reduction in operating costs and a 27.21% maximization of the system’s efficiency while improving the device lifespan by 43.43%. A whole range of benefits of integrating AI into the field of renewable energy were identified by Sahota [26], with the most significant being carbon emission reductions, energy production forecasting, and energy storage.

Contradictory results, however, were generated by the ROBOTS variable, with the results indicating a minimal impact on REN. Thus, we have highlighted that an increase in the number of robots does not bring improvements to renewable energy. Nevertheless, this result is consistent with studies conducted by Bachu [61] and Strubell et al. [40], which emphasized the high energy consumption associated with training and operating AI models.

The Granger causality test is designed specifically to explore potential causal relationships. As illustrated in Figure 4, the application of the causality test revealed two unidirectional causal relationships: of ROBOTS and TOTAL_INV with REN. Additionally, a bidirectional causal relationship was identified between REN and COMPANIES.

This suggests that variations in the number of robots and total investments in AI can predict changes in renewable energy adoption, indicating that these factors are significant drivers of renewable energy trends. Moreover, the bidirectional causality between REN and COMPANIES implies a reciprocal influence, where changes in renewable energy adoption not only affect company behavior but the actions and decisions of companies also impact the level of renewable energy adoption. These findings underscore the complex and interdependent nature of the relationships within the system, highlighting the importance of considering both direct and reciprocal influences when formulating policy recommendations.

5. Conclusions and Policy Implications

This study addresses the critical issue of understanding the relationship between artificial intelligence utilization and renewable energy consumption. By employing advanced analytical techniques such as OLS regression and ENET, this research fills a significant gap in the empirical literature, providing insights into how various predictor variables influence REN.

The findings reveal that according to the OLS regression results, the coefficient for the variable ROBOTS is approximately −0.000174, suggesting that an increase in the number of robots correlates with a slight decrease in REN. Other variables, such as COMPANIES and TOTAL_INV, also demonstrate their respective impacts on REN through their coefficients. In contrast, the ENET estimation identifies a regularization parameter (lambda) of 0.3376, which minimizes the mean squared error. Here, the coefficient for ROBOTS approaches zero, indicating its minimal impact on REN, which differs from the OLS regression results. This contrast highlights the importance of using robust analytical methods like ENET to capture the complex dynamics between AI adoption and sustainability efforts.

The novelty of this research lies in its detailed examination of the effects of AI on renewable energy consumption, particularly in the context of increasing digitization and the environmental concerns that accompany AI technologies. The discrepancies between the OLS regression and ENET results emphasize the necessity for comprehensive analytical frameworks to thoroughly assess the implications of AI for sustainability.

Based on these findings, this study offers several policy recommendations that can inform decision-making in technology, investment, and economic policies. Given the significant coefficients for COMPANIES and TOTAL_INV in both models, policymakers are encouraged to focus on these variables, as they appear to be strong predictors of REN. While the ENET model suggests that investments in robotics may have a minimal direct impact on REN, there is still potential value in supporting initiatives that enhance robotic automation in relevant sectors.

Additionally, the positive coefficient for TOTAL_INV indicates that increased investment levels are associated with greater REN. Therefore, policymakers should promote investments in research and development, infrastructure, and innovation to stimulate economic growth and consequently boost REN. To ensure the effectiveness of these policies, it is crucial to continuously monitor and evaluate their impacts, making necessary adjustments as required. Regular assessments of the model performance and updates to the predictor variables will help maintain the relevance and effectiveness of policy decisions over time.

While this study provides valuable insights into the relationship between artificial intelligence and renewable energy, it is important to acknowledge its limitations. The generalizability of the findings may be limited by the specific dataset and time period analyzed. The results might not be universally applicable across different regions or timeframes. Future research should consider diverse contexts to validate and expand upon these conclusions. The impact of robotics also presents a limitation. The variable ROBOTS shows a minimal influence on REN according to the ENET results, contrasting with the findings from OLS regression. This discrepancy underscores the need for further investigation into the specific effects of robotics on renewable energy, including its potential indirect impacts. Addressing these limitations in future studies could enhance our understanding of AI and robotics’ roles in renewable energy and refine the policy recommendations accordingly.

Following the findings of this study, we aim to explore several research directions to deepen our understanding of the relationship between artificial intelligence, robotics, and renewable energy consumption. Among these, expanding the predictive variables to include policy measures and industrial practices could provide a more comprehensive view of the factors influencing REN. Additionally, a longitudinal analysis could reveal the long-term trends and the evolving impact of AI on renewable energy. Furthermore, exploring various AI technologies, such as machine learning and computer vision, could offer valuable insights into their effects on renewable energy usage. These research directions are not only our objectives but could also be explored by other researchers interested in the intersection between AI and energy sustainability.