A cost-benefit analysis of data processing architectures for the smart grid

Akshay Uttama Nambi Srirangam Narashiman, Matteo Vasirani, R. Venkatesha Prasad, Karl AbererWiMobCity@MobiHoc 2014 · DOI

TL;DR

This paper tackles the crucial challenge of selecting the optimal data processing architecture for smart grids by performing a comprehensive cost-benefit analysis of four distinct models: centralized, decentralized, distributed, and hybrid. The authors introduce and model key cost indicators like monetary cost, accuracy, and privacy, providing valuable insights for smart grid designers to make informed decisions based on specific system requirements.

What Problem Does This Paper Address?

The successful implementation of smart grids hinges on robust Information and Communication Technology (ICT) infrastructure. However, a significant hurdle in large-scale smart grid deployments is the lack of clear guidance on which data processing architecture best meets the diverse requirements for managing the massive influx of data generated by smart meters and other grid components. Existing research often focuses on isolated aspects like communication or storage, leaving a gap in understanding the trade-offs between different architectural choices when considering factors like cost, accuracy of data, and user privacy. This paper aims to bridge this gap by providing a holistic cost-benefit analysis of various architectures.

Methods

The authors propose and analyze four hierarchical data processing architectures for smart grids:

  • Centralized: All data processing and storage occur at a single Utility Control Unit (UCU).
  • Decentralized: Data processing and storage are handled by Neighborhood Area Nodes (NANs), which act as intermediate points between Home Area Nodes (HANs) and the UCU.
  • Distributed: Each Home Area Node (HAN) is responsible for its own data processing and storage, with the UCU initiating queries.
  • Hybrid: Both HANs and NANs possess data processing and storage capabilities, offering a blend of distributed and decentralized approaches.

To evaluate these architectures, the paper introduces and models several key cost indicators:

  • Monetary Cost: Encompasses deployment costs (hardware, installation) and operational costs (energy consumption).
  • Accuracy Cost: Measured using Normalized Root Mean Square Error (NRMSE), reflecting how well original data can be restored from aggregated data.
  • Privacy Cost: Quantified using Shannon entropy, indicating the level of obfuscation of individual customer data.

The analysis considers realistic urban and rural deployment scenarios and incorporates data aggregation techniques (specifically time-wise aggregation) to explore the trade-offs between accuracy and privacy.

Datasets

The paper utilizes data from the REDD initiative [18] for evaluating accuracy and privacy costs. This dataset is described as a public dataset for energy disaggregation research. The authors also refer to Electric Power Research Institute (EPRI) [16] surveys for population density and household data to model urban and rural environments.

Key Insights and Ideas

  • Communication is King (of Cost): Across all architectures, communication emerges as the most significant factor contributing to energy consumption. Reducing communication needs is paramount for energy efficiency.
  • The Accuracy-Privacy Trade-off is Real: Data aggregation, while reducing communication and enhancing privacy, inherently leads to a decrease in data accuracy. The degree of this trade-off is directly influenced by the granularity of aggregation.
  • Centralized is Simple, but Not Scalable: Centralized architectures offer low monetary costs and good accuracy but suffer from scalability issues and are the least privacy-preserving due to the single point of failure and data concentration.
  • Distributed is Private, but Pricey: Distributed architectures offer the highest level of privacy by keeping data local but come with the highest monetary cost due to equipping every HAN with processing and storage capabilities.
  • Hybrid Offers a Balanced Approach: Hybrid architectures aim to strike a balance by distributing processing and storage between HANs and NANs, leading to lower monetary costs than fully distributed systems, albeit with a moderate impact on accuracy due to aggregation at NANs. They are also noted as being more energy-efficient.

Key Results

  • Energy Consumption: Centralized architectures consume significantly more energy, especially in urban environments with a large number of HANs. Hybrid architectures demonstrate the lowest energy consumption.
  • Bandwidth Requirements: Distributed and hybrid architectures require significantly less bandwidth between HANs and NANs due to reduced transmissions. However, NAN-to-UCU bandwidth needs increase with the number of HANs, particularly in urban settings.
  • Storage: Centralized architectures require the most storage as all data is consolidated at the UCU. Other architectures have much lower and more manageable storage needs, with distributed architectures showing the lowest requirement.
  • Monetary Cost: Centralized architectures have the lowest deployment cost. However, when considering operational costs, hybrid architectures become more cost-effective due to their energy efficiency. Distributed architectures have the highest deployment cost due to equipping every HAN.
  • Accuracy: Higher data aggregation granularity directly leads to lower accuracy (higher NRMSE). Centralized and distributed architectures, which don't employ data aggregation, maintain higher accuracy.
  • Privacy: Centralized architectures are the least privacy-preserving. Distributed architectures offer the highest privacy. Decentralized and hybrid architectures fall in between, with privacy levels influenced by the aggregation granularity at NANs.

Limitations and Future Work

  • Disaggregation Algorithm Simplification: The paper acknowledges that the disaggregation algorithm used for accuracy evaluation is a simplified one, focusing on the trade-off rather than achieving optimal disaggregation. Future work could explore more sophisticated disaggregation techniques.
  • Cost Modeling Assumptions: The cost modeling relies on specific functions for storage, communication, and processing costs, which are based on available data. The actual costs might vary depending on specific hardware choices and market conditions.
  • Generic Applicability: While the models are presented as generic, their direct application to any smart grid deployment would require careful calibration of parameters and consideration of specific network topologies and application requirements.
  • Further Exploration of Hybrid Architectures: The paper suggests that hybrid architectures offer a promising balance, and further research could delve deeper into optimizing the distribution of processing and storage between HANs and NANs within this model.
  • Integration of Other Cost Factors: The study focuses on monetary cost, accuracy, and privacy. Future work could incorporate other critical factors like security, reliability, and latency into the cost-benefit analysis.

Summary updated

BibTeX
@inproceedings{narashiman2014cost,
  author = {Akshay Uttama Nambi Srirangam Narashiman and Matteo Vasirani and R. Venkatesha Prasad and Karl Aberer},
  title = {{A cost-benefit analysis of data processing architectures for the smart grid}},
  booktitle = {WiMobCity@MobiHoc},
  year = {2014},
  doi = {10.1145/2633661.2633665},
}

Loading related content...

Figure