Virtual Collection for Distributed Photovoltaic Data: Challenges, Methodologies, and Applications

Leijiao Ge, Tianshuo Du, Changlu Li, Yuanliang Li, Jun Yan, Muhammad Umer RafiqEnergies 2022 · DOI

TL;DR

This paper reviews "virtual collection" for distributed photovoltaic systems (DPVS), a novel approach to gather operational data without extensive physical sensors. It outlines the challenges, methods (similarity analysis, reference station selection, data inference), and applications of this cost-effective and efficient technique, aiming to guide future research and development in the DPVS industry.

What Problem Does This Paper Address?

The rapid growth of distributed photovoltaic systems (DPVS) has led to a significant challenge: collecting comprehensive and reliable operational data. Deploying a sensor for every DPVS unit is prohibitively expensive and logistically complex, especially given the scattered nature of these systems. Furthermore, existing data collection methods are susceptible to transmission blockages and equipment failures. This paper addresses the need for a more efficient and cost-effective solution by exploring "virtual collection," a method that infers data from a limited set of strategically placed "reference power stations" (RPSs).

Methods

The paper breaks down the virtual collection process into three core steps:

  1. Similarity Analysis: This involves identifying DPVS units that share similar characteristics (geographical, meteorological, equipment) within a region. Methods discussed include distance-based approaches (Euclidean, Manhattan, Mahalanobis, DTW) and non-distance-based approaches (Pearson correlation, distance correlation, cosine similarity). The goal is to ensure that data from RPSs is representative of the stations to be virtually collected.
  2. Reference Power Station (RPS) Selection: This is a critical step where the most informative DPVS units are chosen to act as RPSs. The paper categorizes methods into:
  3. Clustering-based algorithms: Partition clustering, hierarchical clustering, density-based clustering, and grid-based clustering are used to group similar DPVS units, with the closest to the cluster centroid selected as RPS.
  4. Optimization-based algorithms: These methods frame RPS selection as a combinatorial optimization problem, often using intelligent algorithms like naturalistic optimization, evolutionary algorithms, and swarm intelligence algorithms. These approaches can be combined with data inference models for better performance evaluation.
  5. Data Inference: Once RPSs are selected, their real-time data is used to estimate the operational data of other DPVS units. This step leverages computational intelligence algorithms. The paper highlights two main categories of models:
  6. Neural network-based models: This includes traditional artificial neural networks, deep neural networks, and specialized architectures like Radial Basis Function (RBF) networks and Wavelet Neural Networks.
  7. Ensemble learning-based models: This encompasses techniques like bagging (e.g., Random Forests), boosting (e.g., XGBoost, LightGBM), and stacking, which combine multiple models to improve prediction accuracy and robustness.

Datasets

Not explicitly discussed in the paper. The paper focuses on a comprehensive review of methodologies and challenges rather than presenting new experimental results with specific datasets.

Key Insights and Ideas

  • Virtual Collection as a Paradigm Shift: The core idea is to move away from ubiquitous sensing towards intelligent inference, significantly reducing costs and complexity.
  • The Importance of Similarity: Accurate virtual collection hinges on selecting RPSs that truly represent the characteristics of the target DPVS units. Poor similarity leads to inaccurate inferences.
  • RPS Selection is Non-Trivial: Unlike standard feature selection, RPS selection needs to consider not only data characteristics but also the impact on the overall system's data estimation accuracy, including temporal and spatial coupling.
  • Data Inference Needs Robustness: Virtual collection requires real-time estimation, demanding data inference models that are more robust and accurate than traditional PV power prediction models, which often focus on future forecasting.
  • Diverse Applications: Virtual collection isn't just about data collection; it has significant implications for anomaly detection, fault diagnosis, and missing data recovery in DPVS.

Key Results

As this is a review paper, it does not present new experimental results. Instead, it synthesizes existing research to:

  • Categorize and summarize various similarity analysis, RPS selection, and data inference methods applicable to DPVS virtual collection.
  • Identify key challenges in implementing virtual collection, such as the need for strict prerequisites and complex data preprocessing.
  • Propose four application scenarios for virtual collection technology in the PV field, highlighting its practical value.

Limitations and Future Work

  • Lack of Specific Datasets: The paper is a review and does not present empirical results on specific datasets, limiting direct quantitative comparisons of the discussed methods.
  • Real-time Accuracy Challenges: The paper notes that achieving high real-time estimation accuracy for virtual collection is challenging due to complex operational environments and the need to account for uncertainty factors.
  • Future Work: The authors suggest further research into virtual collection from the perspective of physical models and providing more practical examples. They also highlight the need to verify the discussed approaches using physical models.

Summary updated

Abstract

In recent years, with the rapid development of distributed photovoltaic systems (DPVS), the shortage of data monitoring devices and the difficulty of comprehensive coverage of measurement equipment has become more significant, bringing great challenges to the efficient management and maintenance of DPVS. Virtual collection is a new DPVS data collection scheme with cost-effectiveness and computational efficiency that meets the needs of distributed energy management but lacks attention and research. To fill the gap in the current research field, this paper provides a comprehensive and systematic review of DPVS virtual collection. We provide a detailed introduction to the process of DPVS virtual collection and identify the challenges faced by virtual collection through problem analogy. Furthermore, in response to the above challenges, this paper summarizes the main methods applicable to virtual collection, including similarity analysis, reference station selection, and PV data inference. Finally, this paper thoroughly discusses the diversified application scenarios of virtual collection, hoping to provide helpful information for the development of the DPVS industry.
BibTeX
@inproceedings{ge2022virtual,
  author = {Leijiao Ge and Tianshuo Du and Changlu Li and Yuanliang Li and Jun Yan and Muhammad Umer Rafiq},
  title = {{Virtual Collection for Distributed Photovoltaic Data: Challenges, Methodologies, and Applications}},
  booktitle = {Energies},
  year = {2022},
  doi = {10.3390/en15238783},
}

Loading related content...

Figure