<p>The advancement in centralized power systems has elevated the power grid into a sophisticated smart grid, emblematic of cyber-physical systems (CPS), susceptible to diverse types of false data injection (FDI) and cyber threats. Among these threats, load frequency control (LFC) systems, crucial for regulating power in tie-lines and ensuring frequency synchronization, are particularly vulnerable to FDI attacks. These attacks pose substantial risks to system continuity, stability, and reliability. Modern power systems consist of multi-power sharing hubs and communication systems integrated with dynamic load demand, such as electric transportation (electric bicycles and cars). This complex mechanism requires a stable power sharing mechanism to meet the load demands of residential, commercial, and charging infrastructure, as well as a robust CPS design that can withstand cyber attacks. This paper introduces a novel control algorithm consisting of a grasshopper optimization algorithm (GOA), a proportional derivative filter (PDF), and a proportional integral (PI). We use the algorithm <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="521_2024_10961_Article_IEq1.gif" Format="GIF" Height="19" Rendition="HTML" Resolution="72" Type="Linedraw" Width="191" /> </InlineMediaObject> <EquationSource Format="TEX">$\text {GOA}-\text {PDF}+(0.75+\text {PI})$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mtext>GOA</mtext> <mo>-</mo> <mtext>PDF</mtext> <mo>+</mo> <mo stretchy="false">(</mo> <mn>0.75</mn> <mo>+</mo> <mtext>PI</mtext> <mo stretchy="false">)</mo> </mrow> </math></EquationSource> </InlineEquation> to optimize the LFC on multiple load deviations. Afterward, a supervisory control and data acquisition system trains an artificial intelligence/machine learning-based model to detect FDI attacks in a multi-area network of centralized renewable energy power systems. Initially, we trained a Levenberg–Marquardt fast neural network (LMFNN) on data related to renewable centralized power generation, frequency aberrations, tie-line power deviations, electrical vehicle recharging, and active power load deviations in both areas. Finally, to find FDI, we compare the LMFNN’s output control signal with the actual output of the plant to find residuals that show FDI attacks. This gives us a remarkable 0.99 regression coefficient score (<i>R</i>), which lets us tell the difference between systems that are under attack and those that are working normally. The efficacy of the proposed technique is demonstrated through simulation models at Matlab 2023b, considering centralized power generation from solar, wind, and thermal power plants. This approach presents a promising solution for improving the resilience of centralized power grids to FDI attacks while maintaining their operational integrity and security.</p>

AI-enabled frequency synchronization control considering FDI attack using metaheuristic algorithm

期刊论文
发表日期：

2025年1月7日

查看全文

Hasnain Ahmad,
Muhammad Majid Gulzar,
Ghulam Mustafa,
Abdul Qayyum Khan,
Salman Habib,
Ijaz Ahmed

摘要

The advancement in centralized power systems has elevated the power grid into a sophisticated smart grid, emblematic of cyber-physical systems (CPS), susceptible to diverse types of false data injection (FDI) and cyber threats. Among these threats, load frequency control (LFC) systems, crucial for regulating power in tie-lines and ensuring frequency synchronization, are particularly vulnerable to FDI attacks. These attacks pose substantial risks to system continuity, stability, and reliability. Modern power systems consist of multi-power sharing hubs and communication systems integrated with dynamic load demand, such as electric transportation (electric bicycles and cars). This complex mechanism requires a stable power sharing mechanism to meet the load demands of residential, commercial, and charging infrastructure, as well as a robust CPS design that can withstand cyber attacks. This paper introduces a novel control algorithm consisting of a grasshopper optimization algorithm (GOA), a proportional derivative filter (PDF), and a proportional integral (PI). We use the algorithm $\text {GOA}-\text {PDF}+(0.75+\text {PI})$ $GOA - PDF + (0.75 + PI)$ to optimize the LFC on multiple load deviations. Afterward, a supervisory control and data acquisition system trains an artificial intelligence/machine learning-based model to detect FDI attacks in a multi-area network of centralized renewable energy power systems. Initially, we trained a Levenberg–Marquardt fast neural network (LMFNN) on data related to renewable centralized power generation, frequency aberrations, tie-line power deviations, electrical vehicle recharging, and active power load deviations in both areas. Finally, to find FDI, we compare the LMFNN’s output control signal with the actual output of the plant to find residuals that show FDI attacks. This gives us a remarkable 0.99 regression coefficient score (R), which lets us tell the difference between systems that are under attack and those that are working normally. The efficacy of the proposed technique is demonstrated through simulation models at Matlab 2023b, considering centralized power generation from solar, wind, and thermal power plants. This approach presents a promising solution for improving the resilience of centralized power grids to FDI attacks while maintaining their operational integrity and security.