Spatial Field Reconstruction and Sensor Selection in Heterogeneous Sensor Networks With Stochastic Energy Harvesting

Pengfei Zhang, Ido Nevat, Gareth W. Peters, François Septier, Michael A. OsborneIEEE TSP 2018 · OpenReview

TL;DR

This paper tackles the challenge of accurately mapping environmental phenomena (like wind speed) using networks of diverse sensors, some of which are low-quality and rely on unpredictable energy harvesting. They introduce a novel method for reconstructing these spatial fields and an efficient algorithm for selecting the best subset of sensors to activate, ensuring both accuracy and cost-effectiveness.

What Problem Does This Paper Address?

Imagine trying to understand a complex environmental field, like wind patterns across a region, using a network of sensors. The problem is that these sensor networks are often "heterogeneous," meaning they have different capabilities. Some are expensive, high-quality sensors (like professional weather stations), while others are cheap, low-quality sensors that might only work when the phenomenon they're measuring reaches a certain intensity (e.g., wind sensors that only activate above a specific speed).

Adding to the complexity, these sensors are often powered by energy harvesting (like solar power), meaning their availability and accuracy can fluctuate based on energy levels. This makes it difficult to get a reliable picture of the entire field. The paper addresses two key challenges arising from this:

  1. Spatial Field Reconstruction: How do we accurately estimate the entire spatial field (e.g., wind speed everywhere) when we have noisy, heterogeneous measurements from different types of sensors, some of which are intermittently available due to energy harvesting?
  2. Sensor Set Selection: Given a user's request for a specific level of accuracy at a particular location, how do we efficiently choose which sensors to activate to meet that requirement while minimizing activation costs?

Methods

The authors propose a two-pronged approach:

  • Spatial Best Linear Unbiased Estimator (S-BLUE) for Field Reconstruction: To tackle the first problem, they develop a low-complexity algorithm based on the S-BLUE. This method is optimal for linear estimators and focuses on minimizing the Mean Squared Error (MSE) of the field reconstruction. It leverages the first two moments of the data, avoiding the need to compute complex posterior distributions, which is crucial for real-time applications.
  • Cross Entropy Method (CEM) for Sensor Selection: For the second problem, they employ the CEM, a stochastic optimization technique. This method is well-suited for solving combinatorial optimization problems like sensor selection, especially when dealing with binary choices (activate or not activate a sensor). The CEM iteratively refines a probability distribution to find an optimal sensor configuration that meets user-defined Quality of Service (QoS) criteria (like maximum allowed error) while minimizing activation costs.

Datasets

The paper uses two types of datasets for evaluation:

  • Synthetic Data: Generated using a Squared Exponential kernel to simulate spatial random fields for both the phenomenon being measured (e.g., wind) and the energy harvesting process.
  • Real-World Data: The Extreme Wind Storms Catalogue (XWS Datasets), an insurance storm surge database. This dataset provides information on storms, including wind speed measurements, and is used to test the algorithm's performance on real-world meteorological events.

Key Insights and Ideas

  • Heterogeneity and Energy Harvesting are Key Challenges: The paper highlights that realistic sensor networks are not uniform. The combination of sensor quality differences and the stochastic nature of energy harvesting significantly complicates data fusion and reconstruction.
  • S-BLUE Offers a Practical Solution for Reconstruction: By focusing on the first two moments, S-BLUE provides an efficient and computationally feasible way to reconstruct spatial fields, even with heterogeneous and noisy data.
  • CEM for Efficient Sensor Selection: The CEM proves to be a powerful tool for solving the combinatorial sensor selection problem, offering a way to balance accuracy requirements with activation costs without resorting to computationally expensive exhaustive searches or potentially suboptimal relaxations.
  • Low-Quality Sensors Can Be Highly Valuable: The simulations demonstrate that even cheap, low-quality sensors, when deployed densely, can significantly improve the accuracy of spatial field reconstruction, especially when augmented with a few high-quality sensors.
  • Activation Threshold Matters: The performance of low-quality sensors is highly dependent on their activation threshold. Lower thresholds allow for more data but can also introduce more noise, while higher thresholds limit data but might provide cleaner measurements.

Key Results

  • Field Reconstruction Accuracy: Simulations show that the S-BLUE algorithm can accurately reconstruct spatial fields. The Root Mean Squared Error (RMSE) decreases significantly as the activation threshold for low-quality sensors is lowered, up to a certain point.
  • Sensor Augmentation Benefits: Adding low-quality sensors to a network of high-quality sensors consistently reduces the overall RMSE in field reconstruction. The paper presents quantitative results showing this improvement across various configurations.
  • Efficient Sensor Selection: The CEM-based sensor selection algorithm converges quickly (within 10 iterations) and achieves a significantly lower cost compared to convex optimization approaches for various QoS levels. This indicates its practical utility for real-time applications.
  • Real-World Performance: The algorithm was tested on the Extreme Wind Storms Catalogue, demonstrating its ability to handle real-world meteorological data.

Limitations and Future Work

  • Independence Assumption: The paper assumes that the energy harvesting process is independent of the monitored physical phenomenon. While this is a practical scenario in many cases (e.g., solar power and wind), the authors acknowledge that in some situations, these processes might be coupled.
  • Known Hyperparameters: The S-BLUE algorithm assumes that the hyperparameters of the Gaussian Process model (mean and covariance functions) are known. In practice, these might need to be estimated, which could add complexity.
  • Future Work: The paper's contributions lay the groundwork for further research in optimizing sensor networks with energy harvesting. Future work could explore more complex dependencies between energy harvesting and the monitored field, investigate adaptive estimation of GP hyperparameters, and extend the sensor selection to more dynamic scenarios.

Summary updated

Abstract

We address the two fundamental problems of spatial field reconstruction and sensor selection in heterogeneous sensor networks. We consider the case where two types of sensors are deployed: the first consists of expensive, high quality sensors; and the second, of cheap low quality sensors, which are activated only if the intensity of the spatial field exceeds a pre-defined activation threshold (e.g., wind sensors). In addition, these sensors are powered by means of energy harvesting and their time varying energy status impacts on the accuracy of the measurement that may be obtained. We then address the following two important problems: (i) how to efficiently perform spatial field reconstruction based on measurements obtained simultaneously from both networks; and (ii) how to perform query based sensor set selection with predictive MSE performance guarantee. To overcome this problem, we solve the first problem by developing a low complexity algorithm based on the spatial best linear unbiased estimator (S-BLUE). Next, building on the S-BLUE, we address the second problem, and develop an efficient algorithm for query based sensor set selection with performance guarantee. Our algorithm is based on the Cross Entropy method which solves the combinatorial optimization problem in an efficient manner. We present a comprehensive study of the performance gain that can be obtained by augmenting the high-quality sensors with low-quality sensors using both synthetic and real insurance storm surge database known as the Extreme Wind Storms Catalogue.
BibTeX
@article{zhang2018spatial,
  author = {Pengfei Zhang and Ido Nevat and Gareth W. Peters and François Septier and Michael A. Osborne},
  title = {{Spatial Field Reconstruction and Sensor Selection in Heterogeneous Sensor Networks With Stochastic Energy Harvesting}},
  journal = {IEEE Trans. Signal Process.},
  year = {2018},
  url = {https://openreview.net/forum?id=RghcmUtdXY},
}
Figure