AI-Based Applications for Renewable Energy and Storage Systems

The integration of renewable energy resources (RES) and energy storage systems (ESS) reduces the dependence on conventional power generation which is largely responsible for carbon emissions and climate change. In addition, the entire power system is usually made observable by installing phasor measurement units (PMUs) and SCADA systems at specific buses. As a result, power system operators can access various measurements such as voltage and current phasors which provide useful information that can be used for monitoring, protection and control. Furthermore, different measurement devices are augmented into the renewable power plants and ESSs. Consequently, collected measurements are fed to the control center to enable power system operators take fast and accurate corrective actions when contingencies are experienced. Meanwhile, the recent developments of artificial intelligence (AI) techniques can be employed to provide better understanding of the inherent dynamic and operational characteristics of a renewable based power system through the large data collected from different locations.

Therefore, the applications of Machine learning and deep learning models such as support vector machine (SVM), Decision trees, recurrent neural networks (RNN), long short-term memory (LSTM), convolutional neural network (CNN), artificial neural network (ANN), and Autoencoders to power systems are tremendously increasing. These techniques have the ability to capture the inherent static and dynamic states of an electrical network when a fault or loss of equipment are experienced as well as the operation performance of RES power plants and ESSs. In addition, the extraction of spatial and temporal distributions of features within the network facilitates taking fast, accurate and informed decisions in the form of classifications and regression analyses. For example, a set of data collected from a power network can be used to train different deep learning models to characterize the required support for the power grid. The developed models can predict the full states of the system from few tens or hundreds of similar datasets without having to build the grid from the scratch as done in time domain simulations. Additionally, transfer learning techniques can be used to solve the problem of insufficient data and changes in operating conditions especially with high penetration levels of the stochastic renewable power generation.

In this context, this APEC research team is focusing and not limited to the following applications in well-designed projects for high RES penetration and ESS toward 100% renewable based power system:

  • Power Frequency Stability Assessment and Control for RES/ESS based power systems

    The intermittent nature of renewable power generation causes frequent changes in operating points due to their reliance on weather conditions. In addition, the occurrence of contingencies such as generator, line and load tripping may result in stressed scenarios where different forms of instabilities can be experienced. In this context, frequency stability represents a crucial aspect of power system operation and planning. It refers to the ability of a power system to maintain steady frequency following a severe disturbance that results in imbalance between generation and load. Therefore, preventive control actions should be taken to avoid possible cascading outages and even complete blackout. To achieve this objective, deep learning techniques such as CNNs, RNNs and LSTM can be employed to provide fast and accurate predictions of the frequency dynamics using collected measurements from phasor measurement units (PMUs) across the entire system. Therefore, the training of these models will be carried out using a dataset that is either collected from real-time measurements or generated through extensive time domain simulations for wide range of operation conditions and penetration levels of renewable energy resources (RES). The predicted power frequency dynamics may be utilized to identify the required load shedding amount and suboptimal operation point of RES to maintain the maximum frequency deviation value within the acceptable operating limits.
  • Voltage Stability Assessment, Control and Enhancement for 100% Renewable/ESS based Power Systems

    Voltage stability refers to the ability of a power system to maintain steady voltages at all buses after being subjected to a disturbance from a given initial operating condition. Voltage instability can be experienced in the form of a cascaded fall or rise of voltage at specific buses. In addition, voltage instability may result in loss of load(s) in an area, or tripping of transmission lines that may be followed by cascading outages or voltage collapse. This phenomenon can be avoided through real-time monitoring of wide-area measurements that provide useful information about system voltage magnitudes and angles. In this context, deep learning models based on CNNs or LSTM can be trained to provide fast predictions of the voltage stability and inform power system operators about the consequences of severe disturbances if proper control actions are not taken. If properly trained and tuned, these models have the ability to predict how the system will respond to specific contingencies based on the learned dynamics. Therefore, deep learning techniques can provide a more thorough overview and understanding of various interactions at the components and system levels. In the literature, different deep learning techniques are applied to address short-term voltage stability (extracting short-term feature from scarce instability data) assessment; long-term voltage stability assessment, TSSC-based SVS margin estimation and cost-sensitive corrective controls, class imbalance between stable and unstable data.
  • Real-Time Power System Inertia Estimation Using Advanced Artificial Intelligence Techniques

    Increased penetration levels of renewable power generation (RPG) reduce inertia levels as a result of the continuous replacement of synchronous generators. Consequently, a higher rate of change of frequency (RoCoF) and larger frequency deviations from nominal operating conditions are expected when disturbances occur. This research work proposes a real-time deep learning-based models for predicting inertia constants of synchronous generators in the presence of renewable power generation (RPG). The trained models provide power grid operators with a fast and accurate monitoring tool that may be utilized for taking corrective measures against possible frequency instabilities. Five variants of recurrent neural network (LSTM, LSTM-MLP, CNN-LSTM, ConvLSTM and GRU) as well as MLP network are used in this research to accomplish this objective. Furthermore, the proposed inertia prediction models are trained using datasets that carry useful information about the real-time operational settings of a power system. Therefore, various measurements such as active power output, mechanical power, and rate-of-change-of-frequency (RoCoF) of a synchronous generator are used along with output power from solar PV power plants. The root means squared errors (RMSE), means squared errors (MSE) and mean absolute errors (MAE) are computed for the training, validation, and test datasets. Results show all six networks accurately estimate system inertia. LSTM model outperform other models on the three evaluation metrics. The proposed method can estimate one sequence of inertia (i.e., one time-step) in a CPU environment within a time span of 2 - 5 ms. As a result, the proposed approach is computationally efficient and fast.

For the aforementioned applications, the efficacy of Machine learning / deep learning (ML/DL) models lies in the efficient data preparation. Since models are built as networks of graphs with high shareable features that are akin to high dimensional complex interconnections of power grid, they can learn the nonlinearity phenomena of power systems from well prepared data beyond mathematical representation.

Developed AI Software/Tools

Stability Assessment, Visualization and Enhancement (SAVE) Software

SAVE is a novel Artificial Intelligence (AI) software (for the power system operator) that facilitates effective power system stability assessment, visualization, and enhancement. SAVE software can utilize real time measurements and full system observability to reliably predict the power system stability in the presence of uncertainty resulting from renewable power generation and load demand. As its core, SAVE software applies advanced artificial intelligence, deep learning and system identification techniques for power system stability assessment. They demonstrated superior performance in terms of capturing the inherent dynamic characteristics of a power grid and thus can be used to provide fast and accurate predictions for small signal and transient stabilities of power grid. SAVE software largely relies on collected data to train and test power system stability classifier which will enable operators to identify threatening contingencies and prepare the proper control actions accordingly. In addition, it is supported with advanced deterministic methods for power system stability, dynamic mode decomposition technique (DMD) and combined with a data analytics based classification engine to precisely estimate the domain of stability and to predict the transient stability. Furthermore, a visualization system (e.g., a dashboard) has been developed to support human-in-the-loop classification, diagnosis of the transmission system events, and the visualization of system stability.

SAVE software provides also a critical feature of great significance for safe and reliable operation of power systems. The presence of poorly-damped low-frequency modes of oscillations can limit the amount of power transfer over transmission lines. In addition, they can cause cascaded tripping of synchronous generators when small perturbations like load changes are experienced. Therefore, the overall system performance may be severely jeopardized as a result of these modes. This problem has been traditionally addressed through modal analysis that is based on offline study and linearized model about an equilibrium point. Nevertheless, this approach fails to address the issue of changing operating conditions in real-time especially in the presence of highly intermittent renewable energy resources. Therefore, SAVE software employs advanced system identification technique that is dynamic mode decomposition (DMD) to provide accurate estimates of these modes in online settings. DMD was successfully applied to various configurations of the TRANSCO network and IEEE benchmark test systems.

The SAVE software is largely programmed using python-language which is an open-source programming language and facilitates training and evaluating advanced types of deep learning models. A visualization dashboard is also designed to provide a friendly interface where users can interact with the main engine which runs different routines for assessing the stability of TRANSCO’s electrical network. Therefore, the user can upload static and dynamic models of TRANSCO’s network, simulate and visualize contingencies, train and evaluate neural network models and run dynamic mode decomposition (DMD) to evaluate the overall damping of low-frequency modes of oscillations.

This research project is carried out in collaboration with external partners Abu Dhabi Transmission and Dispatch Company (TRANSCO), UAE and Manitoba Hydro International (MHI), Canada (MHI, Khalifa University, & TRANSCO Develop SAVE Software | Manitoba Hydro International). This industrial-academic research collaboration resulted in developing a final product (SAVE Software), publishing 39 high quality journal papers (Q1, Mostly IEEE Transactions), two US Patents, submitted 11 Journal papers (under review) and graduated high calibers MSc and PhD students.

“SAVE Tool is currently being used by TRANSCO planning engineers to evaluate the overall dynamic performance of the TRANSCO network when futuristic expansion plans are anticipated. In addition, SAVE tool was successful put in operation at the control center to assist study engineers and operators for assessing the stability of power grid with renewable energy integration”

KU research team received the First place in the R&D and Innovation Award category for the Universities and Research Centers from the Ministry of Energy and Infrastructure for the developed software ”Stability Assessment, Visualization and Enhancement (SAVE)” in Feb. 2023

For more details about the advanced version of SAVE software, please contact Prof. Mohamed El Moursi, Deputy Director of Advanced Power and Energy Center (APEC), (E-mail: )