Real-time Power System State Estimation and Forecasting via Deep Neural Networks (1811.06146v2)

Abstract: Contemporary power grids are being challenged by rapid voltage fluctuations that are caused by large-scale deployment of renewable generation, electric vehicles, and demand response programs. In this context, monitoring the grid's operating conditions in real time becomes increasingly critical. With the emergent large scale and nonconvexity however, the existing power system state estimation (PSSE) schemes become computationally expensive or yield suboptimal performance. To bypass these hurdles, this paper advocates deep neural networks (DNNs) for real-time power system monitoring. By unrolling an iterative physics-based prox-linear solver, a novel model-specific DNN is developed for real-time PSSE with affordable training and minimal tuning effort. To further enable system awareness even ahead of the time horizon, as well as to endow the DNN-based estimator with resilience, deep recurrent neural networks (RNNs) are also pursued for power system state forecasting. Deep RNNs leverage the long-term nonlinear dependencies present in the historical voltage time series to enable forecasting, and they are easy to implement. Numerical tests showcase improved performance of the proposed DNN-based estimation and forecasting approaches compared with existing alternatives. In real load data experiments on the IEEE 118-bus benchmark system, the novel model-specific DNN-based PSSE scheme outperforms nearly by an order-of-magnitude the competing alternatives, including the widely adopted Gauss-Newton PSSE solver.

• The paper introduces novel deep neural network architectures for real-time power system state estimation and forecasting.
• The proposed deep prox-linear network for state estimation achieves significantly higher accuracy than traditional methods on large power systems.
• Deep recurrent neural networks effectively forecast power system states, enhancing real-time grid monitoring and situational awareness.

Real-time Power System State Estimation and Forecasting via Deep Neural Networks

The paper "Real-time Power System State Estimation and Forecasting via Deep Neural Networks" introduces innovative schemes for power system monitoring using deep learning architectures, specifically focusing on Power System State Estimation (PSSE) and power system state forecasting. The authors address critical challenges faced by modern power grids due to unpredictable voltage fluctuations caused by the rapid adoption of renewable energy sources, electric vehicles, and demand response programs.

Overview of Proposed Methodology

The research advocates for model-specific deep neural networks (DNNs) in performing real-time PSSE. By unrolling an iterative physics-based prox-linear solver, the authors design a novel DNN architecture tailored to handle real-time power system monitoring efficiently. This approach is contrasted with existing methods that rely heavily on optimization techniques, such as weighted least-squares (WLS) and least-absolute-value (LAV) estimators, which can become computationally demanding in the context of large-scale, nonconvex optimization.

To complement PSSE, the paper leverages deep recurrent neural networks (RNNs) for power system state forecasting, capitalizing on the ability of RNNs to manage long-term nonlinear dependencies associated with historical voltage time series. The combination of DNNs for PSSE and RNNs for forecasting supports enhanced system observability and enables situational awareness ahead of the actual time horizon.

Numerical Results and Performance Assessment

Extensive numerical experiments were conducted using the IEEE 57-bus and 118-bus benchmark systems. The proposed deep prox-linear net for PSSE considerably outperforms traditional estimators such as the Gauss-Newton method, with the former boasting nearly an order-of-magnitude superior accuracy in the IEEE 118-bus system tests. Specifically, the prox-linear net showcases impressive robustness against the nonconvexities typical of the power system state estimation problem.

Furthermore, the research demonstrates the efficacy of deep RNNs compared to first-order vector auto-regressive (VAR) models and single-hidden-layer feedforward neural networks used in state forecasting. The deep RNNs achieve reduced normalized root mean-square errors (RMSE), highlighting their proficiency in capturing complex temporal patterns in voltage data.

Implications and Future Research Directions

The implications of this research are profound both in practical automation within energy systems and theoretical advancements in applying deep learning to large-scale, complex cyber-physical systems. The proposed frameworks promise improvements in real-time processing capabilities, thus facilitating more responsive and adaptable energy management systems within smart grids. Moreover, they stage a promising ground for extending functionality to distribution networks and investigating adaptive algorithms that could accommodate dynamically changing environments.

Future research could delve into further tailoring these deep learning models to target specific grid configurations or operational conditions. Additionally, exploring online learning paradigms to refine recurrent networks with real-time data updates could enhance robustness amid fluctuating system parameters or unforeseen events.

This paper contributes significantly to the field, presenting a cohesive strategy marrying physics-based solver insights with advanced neural network architectures in promoting resilient, scalable, and efficient power system monitoring solutions.