the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 25 Sep 2023
Carbon Monitor Power - Simulators (CMP-SIM v1.0) across countries: a data-driven approach to simulate daily power demand
Abstract. The impact of climate change on power demand has become increasingly significant, with changes in temperature, relative humidity, and other climate variables affecting cooling and heating demand for households and industries. Accurately predicting power demand is crucial for energy system planning and management. It is also crucial to understand the evolution of power demand to estimate the amount of CO2 emissions released into the atmosphere, allowing stakeholders to make informed plans to reduce emissions and adapt to the impacts of climate change. Artificial intelligence techniques have been used to investigate energy demand-side responses to external factors at various scales in recent years. However, few have explored the impact of climate and weather variability on power demand. This study proposes a data-driven approach to model daily power demand provided by the Carbon Monitor Power project by combining climate variables and human activity indices as predictive features. Our investigation spans the years 2020 to 2022 and focuses on eight countries or groups of countries selected to represent different climates and economies, accounting for over 70 % of global power consumption. These countries include Australia, Brazil, China, the European Union (EU), India, Russia, South Africa, and the United States. We assessed various machine-learning regressors to simulate daily power demand at the national scale. For countries within the EU, we extended the analysis to one group of countries. We evaluated the models based on key evaluating metrics: coefficient of determination (R2), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Median Absolute Error (MedAE). We also used the models to identify the most influential variables that impact power demand and apprehend their relationship with it. Our findings provide insight into variations in important predictive features among countries, along with the role played by distinct climate variables and indicators of the level of economic activity, such as weekends and working days, vacations and holidays, and the influence of COVID-19.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(2579 KB)
-
Supplement
(27078 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2579 KB) - Metadata XML
-
Supplement
(27078 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-1313', Matteo De Felice, 23 Oct 2023
The article covers a very interesting topic and the authors try to assess a wide range of models in multiple regions. However, the article presents two main flaws:
1. The authors do not explain what the goal of this power demand model would be, given the challenges posed by this task, its target use is fundamental to understand the quality (and the usefulness) of the results
2. There is no mention to many works on the link between electricity demand and meteorological factors, as for example in the Copernicus Climate Change Service ECEM project and papers like (this is just an example) https://rmets.onlinelibrary.wiley.com/doi/10.1002/met.1858, https://iopscience.iop.org/article/10.1088/1748-9326/11/12/124025/meta, https://iopscience.iop.org/article/10.1088/1748-9326/aa69c6/meta
In addition, I would highlight a few issues that should be addressed:
1. The authors should use a simple model as a baseline (for example a linear regression) to show the value of using more complex methodologies to model the power demand. In other words, showing the added value of non-linearity or ensemble approaches.
2. The authors define the power demand as the total generation, assuming no cross-border exchanges of electricity that happen in many of the selected regions (e.g., exchanges between US and Canada). I think that defining this methodology as "daily power demand simulation" is a bit stretched, perhaps it would be more correct to change the title of the paper to "simulate daily power generation".
3. TOY is not correctly coded in the methodology, using a linear factor put 1st January and 31st December at the opposites, while they are actually consecutive. I would suggest using a sinusoidal function.Citation: https://doi.org/10.5194/egusphere-2023-1313-RC1 -
AC1: 'Reply on RC1', Léna Gurriaran, 11 Dec 2023
Thank you for your thoughtful and constructive feedback. We appreciate the time you have taken to review our manuscript. Below is a point-by-point response to your comments:
- Clear Definition of Objective:
While we explained what our goal is with the electricity demand models in the last part of our manuscript (Perspective), we recognize the importance of clearly defining the goal of our models from the beginning of the article. In the next version, we will add a dedicated paragraph to the introduction that explicitly states our primary goal. Our objective is to project future electricity demand using models that incorporate IPCC projections of climate variables (CMIP6 models) under different scenarios (Shared Socioeconomic Pathways). This approach allows us to identify potential trends in electricity demand, with a particular focus on the impact of climate change on heating and cooling demand. Finally, we attempt to calculate the associated CO2 emissions to address the question of whether climate change is significantly influencing electricity demand, potentially triggering a feedback loop on global temperatures.
- Incorporating Additional References:
Thank you for suggesting additional references. We plan to expand our bibliography, particularly in the introduction, to provide a more comprehensive overview of our study's contribution in the revised version of the article. While our initial literature review focused primarily on machine learning load forecasting, we recognize the importance of including references on the relationship between climate/weather variability and electricity demand.
The references you suggested include references to a multi-linear model for demand simulation (Bloomfield et al., 2016; Bloomfield et al., 2019). We intend to include these references to expand our literature review. In addition, we will include Dubus et al, 2021 (https://doi.org/10.31223/X5MM06) on the C3S Energy service. We also plan to include two additional papers that specifically address the use of GAM models and boosting approaches for load forecasting: Fan and Hyndman, 2012 (https://doi.org/10.1109/PESGM.2012.6345304) and Hong et al, 2016 (https://doi.org/10.1016/j.ijforecast.2016.02.001). These latter papers provide evidence for the superiority of GAM and nonparametric approaches in the context of load forecasting. These additional references also enable us to address your subsequent comment regarding the necessity of a simple baseline model.
- Simple Baseline Model:
Based on these references, we argue that the existing evidence demonstrates the superior performance of GAM and machine learning approaches compared to multi-linear models. This leads us to question the necessity of including a benchmark study in our article to establish the added value of nonparametric approaches. However, recognizing the importance of a comprehensive evaluation, we are open to including a benchmark section in the revised version of the article if you deem it necessary.
We also want to emphasize that our chosen approach, particularly with GAMs, is designed for flexibility and ease of inclusion of multiple variables. The interpretability of GAMs, represented as a sum of spline functions, allows for a straightforward understanding of each explanatory feature. For other machine learning models, we use interpretability methods to automatically identify significant variables. This, in turn, facilitates the exclusion of climatic factors that do not contribute meaningfully to our electricity demand forecasting studies. In addition, our approach minimizes the need for manual tuning compared to multi-linear models, providing efficiency and automation.
- Terminology Adjustment:
We acknowledge your suggestion to change "daily power demand simulation" to "simulate daily power generation" and will incorporate this adjustment in the revised version.
- TOY Variables:
We argue that the coding of TOY variables is correct for GAM models because it automatically generates cyclic splines. While sinusoidal functions could improve accuracy for other machine learning models, our current approach with tree-based models efficiently handles these variables thanks to its ability to easily generate thresholds, thus allowing the first and last days of the year to be placed in the same category.
Citation: https://doi.org/10.5194/egusphere-2023-1313-AC1
-
AC1: 'Reply on RC1', Léna Gurriaran, 11 Dec 2023
-
RC2: 'Comment on egusphere-2023-1313', Giacomo Falchetta, 31 Oct 2023
Many thanks for the opportunity to review this very interesting, well written, and comprehensively presented paper.
While align with all the comments/criticism pointed out by Referee 1, in particular on the necessity of revising the terminology (e.g., generation, and not demand) and of testing simpler models to show the value added of non-parametric statistical modelling, I a couple of additional comments to add.
Fist of all, while certainly significant and novel, the study should cite previous similar papers, e.g. (just an example, there is likely more) https://www.sciencedirect.com/science/article/pii/S0142061518336196, which are missing from the review of the literature in the first part of the paper. The authors should then better emphatise their contribution compared to previous large-scale energy production/generation demand studies.
Moreover, the authors validate the model using daily resolution power generation. I think a crucial and valuable addition to demonstrate the extent to which the model and output data can be used for plannning purposes would be to also evaluate the model error in each country/region in terms of weekly/monthly/seasonal peak. This is because the peak load (maximum value)'s magnitude and modelling error are of great importance if the data is used in future studies and/or for planning and policy support purposes.
Finally, and relatedly, it would be interesting if the authors could explicitly account for the availability and use of cooling and heating technologies in each country, as these are strongly affecting the relation between meteorological variables and energy consumption, see https://www.nature.com/articles/s41598-023-31469-z
Citation: https://doi.org/10.5194/egusphere-2023-1313-RC2 -
AC2: 'Reply on RC2', Léna Gurriaran, 11 Dec 2023
Thank you for your thorough review and positive feedback on our paper. We appreciate your valuable insights and would like to address the additional comments you provided. Here is a point-by-point response:
- Citation of previous similar papers
As mentioned in our response to the previous comment, we are committed to improving the literature review in the Introduction to better contextualize our work and highlight its contribution. Specifically, we will include a paragraph illustrating how nonparametric models streamline automation and application across countries. To our knowledge, there is a gap in global studies utilizing such nonparametric methods. Existing data-driven studies rely on multilinear models, as exemplified in the article you cited.
- Model Validation with extremes
In terms of evaluating the performance of our model under extreme conditions, we acknowledge its current limitations in dealing with extremes. In our future studies, we plan to explore specialized models designed for extreme conditions. Recent works, including Fasiolo et al. (2021, https://doi.org/10.1080/01621459.2020.1725521), Velthoen et al. (2022, https://doi.org/10.48550/arXiv.2103.00808), and Gnecco et al. (2023, https://doi.org/10.48550/arXiv.2201.12865), have explored novel machine learning approaches to improve the understanding and prediction of extreme quantiles. We will discuss these limitations and potential advances in the Discussion section (5.2), and explicitly mention in the Perspectives section that we aim to develop models tailored to extremes.
- Accounting for cooling and heating technologies
We are currently working on multi-country modeling to address data gaps, particularly in regions with limited air conditioning infrastructure, such as Europe compared to Japan or the US. This involves taking climate-energy demand relationships from one country and applying them to another. For example, we plan to simulate European electricity demand using the electricity-demand-climate relationship observed in Japan above the cooling threshold. In addition, our ongoing project involves refining the projection aspect by incorporating finer spatial resolution data. This includes integrating more socio-economic predictors, including demographic characteristics and building characteristics such as insulation and exposure.
Citation: https://doi.org/10.5194/egusphere-2023-1313-AC2
-
AC2: 'Reply on RC2', Léna Gurriaran, 11 Dec 2023
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-1313', Matteo De Felice, 23 Oct 2023
The article covers a very interesting topic and the authors try to assess a wide range of models in multiple regions. However, the article presents two main flaws:
1. The authors do not explain what the goal of this power demand model would be, given the challenges posed by this task, its target use is fundamental to understand the quality (and the usefulness) of the results
2. There is no mention to many works on the link between electricity demand and meteorological factors, as for example in the Copernicus Climate Change Service ECEM project and papers like (this is just an example) https://rmets.onlinelibrary.wiley.com/doi/10.1002/met.1858, https://iopscience.iop.org/article/10.1088/1748-9326/11/12/124025/meta, https://iopscience.iop.org/article/10.1088/1748-9326/aa69c6/meta
In addition, I would highlight a few issues that should be addressed:
1. The authors should use a simple model as a baseline (for example a linear regression) to show the value of using more complex methodologies to model the power demand. In other words, showing the added value of non-linearity or ensemble approaches.
2. The authors define the power demand as the total generation, assuming no cross-border exchanges of electricity that happen in many of the selected regions (e.g., exchanges between US and Canada). I think that defining this methodology as "daily power demand simulation" is a bit stretched, perhaps it would be more correct to change the title of the paper to "simulate daily power generation".
3. TOY is not correctly coded in the methodology, using a linear factor put 1st January and 31st December at the opposites, while they are actually consecutive. I would suggest using a sinusoidal function.Citation: https://doi.org/10.5194/egusphere-2023-1313-RC1 -
AC1: 'Reply on RC1', Léna Gurriaran, 11 Dec 2023
Thank you for your thoughtful and constructive feedback. We appreciate the time you have taken to review our manuscript. Below is a point-by-point response to your comments:
- Clear Definition of Objective:
While we explained what our goal is with the electricity demand models in the last part of our manuscript (Perspective), we recognize the importance of clearly defining the goal of our models from the beginning of the article. In the next version, we will add a dedicated paragraph to the introduction that explicitly states our primary goal. Our objective is to project future electricity demand using models that incorporate IPCC projections of climate variables (CMIP6 models) under different scenarios (Shared Socioeconomic Pathways). This approach allows us to identify potential trends in electricity demand, with a particular focus on the impact of climate change on heating and cooling demand. Finally, we attempt to calculate the associated CO2 emissions to address the question of whether climate change is significantly influencing electricity demand, potentially triggering a feedback loop on global temperatures.
- Incorporating Additional References:
Thank you for suggesting additional references. We plan to expand our bibliography, particularly in the introduction, to provide a more comprehensive overview of our study's contribution in the revised version of the article. While our initial literature review focused primarily on machine learning load forecasting, we recognize the importance of including references on the relationship between climate/weather variability and electricity demand.
The references you suggested include references to a multi-linear model for demand simulation (Bloomfield et al., 2016; Bloomfield et al., 2019). We intend to include these references to expand our literature review. In addition, we will include Dubus et al, 2021 (https://doi.org/10.31223/X5MM06) on the C3S Energy service. We also plan to include two additional papers that specifically address the use of GAM models and boosting approaches for load forecasting: Fan and Hyndman, 2012 (https://doi.org/10.1109/PESGM.2012.6345304) and Hong et al, 2016 (https://doi.org/10.1016/j.ijforecast.2016.02.001). These latter papers provide evidence for the superiority of GAM and nonparametric approaches in the context of load forecasting. These additional references also enable us to address your subsequent comment regarding the necessity of a simple baseline model.
- Simple Baseline Model:
Based on these references, we argue that the existing evidence demonstrates the superior performance of GAM and machine learning approaches compared to multi-linear models. This leads us to question the necessity of including a benchmark study in our article to establish the added value of nonparametric approaches. However, recognizing the importance of a comprehensive evaluation, we are open to including a benchmark section in the revised version of the article if you deem it necessary.
We also want to emphasize that our chosen approach, particularly with GAMs, is designed for flexibility and ease of inclusion of multiple variables. The interpretability of GAMs, represented as a sum of spline functions, allows for a straightforward understanding of each explanatory feature. For other machine learning models, we use interpretability methods to automatically identify significant variables. This, in turn, facilitates the exclusion of climatic factors that do not contribute meaningfully to our electricity demand forecasting studies. In addition, our approach minimizes the need for manual tuning compared to multi-linear models, providing efficiency and automation.
- Terminology Adjustment:
We acknowledge your suggestion to change "daily power demand simulation" to "simulate daily power generation" and will incorporate this adjustment in the revised version.
- TOY Variables:
We argue that the coding of TOY variables is correct for GAM models because it automatically generates cyclic splines. While sinusoidal functions could improve accuracy for other machine learning models, our current approach with tree-based models efficiently handles these variables thanks to its ability to easily generate thresholds, thus allowing the first and last days of the year to be placed in the same category.
Citation: https://doi.org/10.5194/egusphere-2023-1313-AC1
-
AC1: 'Reply on RC1', Léna Gurriaran, 11 Dec 2023
-
RC2: 'Comment on egusphere-2023-1313', Giacomo Falchetta, 31 Oct 2023
Many thanks for the opportunity to review this very interesting, well written, and comprehensively presented paper.
While align with all the comments/criticism pointed out by Referee 1, in particular on the necessity of revising the terminology (e.g., generation, and not demand) and of testing simpler models to show the value added of non-parametric statistical modelling, I a couple of additional comments to add.
Fist of all, while certainly significant and novel, the study should cite previous similar papers, e.g. (just an example, there is likely more) https://www.sciencedirect.com/science/article/pii/S0142061518336196, which are missing from the review of the literature in the first part of the paper. The authors should then better emphatise their contribution compared to previous large-scale energy production/generation demand studies.
Moreover, the authors validate the model using daily resolution power generation. I think a crucial and valuable addition to demonstrate the extent to which the model and output data can be used for plannning purposes would be to also evaluate the model error in each country/region in terms of weekly/monthly/seasonal peak. This is because the peak load (maximum value)'s magnitude and modelling error are of great importance if the data is used in future studies and/or for planning and policy support purposes.
Finally, and relatedly, it would be interesting if the authors could explicitly account for the availability and use of cooling and heating technologies in each country, as these are strongly affecting the relation between meteorological variables and energy consumption, see https://www.nature.com/articles/s41598-023-31469-z
Citation: https://doi.org/10.5194/egusphere-2023-1313-RC2 -
AC2: 'Reply on RC2', Léna Gurriaran, 11 Dec 2023
Thank you for your thorough review and positive feedback on our paper. We appreciate your valuable insights and would like to address the additional comments you provided. Here is a point-by-point response:
- Citation of previous similar papers
As mentioned in our response to the previous comment, we are committed to improving the literature review in the Introduction to better contextualize our work and highlight its contribution. Specifically, we will include a paragraph illustrating how nonparametric models streamline automation and application across countries. To our knowledge, there is a gap in global studies utilizing such nonparametric methods. Existing data-driven studies rely on multilinear models, as exemplified in the article you cited.
- Model Validation with extremes
In terms of evaluating the performance of our model under extreme conditions, we acknowledge its current limitations in dealing with extremes. In our future studies, we plan to explore specialized models designed for extreme conditions. Recent works, including Fasiolo et al. (2021, https://doi.org/10.1080/01621459.2020.1725521), Velthoen et al. (2022, https://doi.org/10.48550/arXiv.2103.00808), and Gnecco et al. (2023, https://doi.org/10.48550/arXiv.2201.12865), have explored novel machine learning approaches to improve the understanding and prediction of extreme quantiles. We will discuss these limitations and potential advances in the Discussion section (5.2), and explicitly mention in the Perspectives section that we aim to develop models tailored to extremes.
- Accounting for cooling and heating technologies
We are currently working on multi-country modeling to address data gaps, particularly in regions with limited air conditioning infrastructure, such as Europe compared to Japan or the US. This involves taking climate-energy demand relationships from one country and applying them to another. For example, we plan to simulate European electricity demand using the electricity-demand-climate relationship observed in Japan above the cooling threshold. In addition, our ongoing project involves refining the projection aspect by incorporating finer spatial resolution data. This includes integrating more socio-economic predictors, including demographic characteristics and building characteristics such as insulation and exposure.
Citation: https://doi.org/10.5194/egusphere-2023-1313-AC2
-
AC2: 'Reply on RC2', Léna Gurriaran, 11 Dec 2023
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 315 | 140 | 23 | 478 | 30 | 13 | 20 |
- HTML: 315
- PDF: 140
- XML: 23
- Total: 478
- Supplement: 30
- BibTeX: 13
- EndNote: 20
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Léna Gurriaran
Yannig Goude
Katsumasa Tanaka
Biqing Zhu
Zhu Deng
Xuanren Song
Philippe Ciais
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(2579 KB) - Metadata XML
-
Supplement
(27078 KB) - BibTeX
- EndNote
- Final revised paper