Abstract
A multivariate density forecast model based on deep learning is designed in this paper to forecast the joint cumulative distribution functions (JCDFs) of multiple security margins in power systems. Differing from existing multivariate density forecast models, the proposed method requires no a priori hypotheses on the distribution of forecasting targets. In addition, based on the universal approximation capability of neural networks, the value domain of the proposed approach has been proven to include all continuous JCDFs. The forecasted JCDF is further employed to calculate the deterministic security assessment index evaluating the security level of future power system operations. Numerical tests verify the superiority of the proposed method over current multivariate density forecast models. The deterministic security assessment index is demonstrated to be more informative for operators than security margins as well.
Introduction
Modern power grids are undergoing a significant transformation with the integration of volatile renewable energy sources and power electronic devices. This shift introduces unprecedented levels of uncertainty, making traditional model-driven security assessment methods increasingly inadequate. While data-driven point estimation offers some relief, it often fails to capture the probabilistic nature and interdependencies of various security margins. This paper proposes a novel multivariate density forecast model, JDAN-NFN, to address these challenges by forecasting joint cumulative distribution functions (JCDFs) without requiring prior distributional assumptions.
Methodology: The JDAN-NFN Framework
The proposed framework consists of two primary neural network components: the Network Forecast Network (NFN) and the Joint Distribution Approximation Network (JDAN).
Architecture
The NFN utilizes Long Short-Term Memory (LSTM) layers to process historical time-series data, such as bus voltages and power flows. Instead of outputting security margins directly, the NFN predicts the optimal weights ($W^+$) and biases ($B$) for the subsequent JDAN.
The JDAN is specifically engineered to approximate the JCDF. Unlike standard neural networks, JDAN incorporates structural constraints to ensure the output adheres to the mathematical properties of a distribution function (e.g., monotonicity and range [0, 1]). It features parallel units for individual margins, normalization layers, a fusion layer, and a critical coupling layer that captures the dependencies between different security margins.
Deterministic Security Assessment Index
To translate complex probabilistic forecasts into actionable insights, the authors introduce the Deterministic Security Assessment Index ($\Omega$). This index integrates the forecasted JCDF to provide a single probability value representing the likelihood that the system will remain within safe operating limits over a future horizon (e.g., 15 minutes). This provides operators with a more intuitive measure of risk than raw security margins.
Experimental Results
The model was evaluated using a modified New England 39-bus system featuring two 400MW wind farms to introduce realistic volatility. Data was generated via Monte-Carlo sampling and continuation power flow simulations, resulting in over 15,000 samples.
Performance and Efficiency
The JDAN-NFN model demonstrated exceptional accuracy, with reliability evaluation errors between 0.62% and 1.22%. This significantly outperforms traditional methods like Multivariate Kernel Density Estimation (MKDE) and Copula-based models, which exhibited errors exceeding 7%. Furthermore, the computational efficiency is noteworthy; the average time to calculate $\Omega$ is approximately 0.0492 seconds, making it highly suitable for real-time online assessment.
Risk Awareness
A key finding is that the $\Omega$ index provides a more nuanced view of grid security than raw observations. It can distinguish between scenarios that appear identical in point estimates but possess different underlying risk profiles due to forecasted volatility. This allows for proactive rather than reactive grid management.
Conclusion
The JDAN-NFN framework represents a significant advancement in power system security assessment. By providing a distribution-free, multivariate density forecast, it offers grid operators a robust and efficient tool for managing the uncertainties of modern power systems. Future work will explore scaling the model to larger national grids and refining the coupling formulations to capture even more complex variable dependencies.
Summary updated