Detection of False Data Injection Attacks (FDIA) Targeting Distribution Transformers in Smart Grids Using Hybrid Artificial Intelligence Methods

Recep AKKAN; Osman Can ÇETLENBİK

doi:10.2478/ias-2026-0011

.blurhash-client-img { display: none !important; }

Detection of False Data Injection Attacks (FDIA) Targeting Distribution Transformers in Smart Grids Using Hybrid Artificial Intelligence Methods

Journal of Information Assurance and Security

Volume 21 (2026): Issue 4 (July 2026)

By: Recep AKKAN and Osman Can ÇETLENBİK

Open Access

|Jul 2026

Figures & Tables

IEC 61850-based intelligent substation communication architecture illustrating the interaction between IEDs, SCADA infrastructure, and FDIA attack surfaces through vulnerable GOOSE/MMS communication channels

Temporal variations in voltage, current, and active power measurements at Bus 69 under normal operating conditions and FDIA attack scenarios, where the shaded red region indicates the active attack interval

Receiver Operating Characteristic (ROC) curves comparing the FDIA detection performance of LSTM, Isolation Forest, and One-Class SVM models on the IEEE 118-bus dataset

Confusion matrices obtained from the evaluation of LSTM, Isolation Forest, and One-Class SVM models on the 500-sample test dataset containing 100 FDIA instances and 400 normal operating samples

Comparative performance analysis of the evaluated FDIA detection models using Accuracy, Recall, Precision, F1-score, and AUC-ROC metrics on the IEEE 118-bus test system

LSTM Performance Stability Across 5 Independent Runs

Run	Recall (%)	F1-Score	AUC-ROC
Run 1	97.0	0.965	0.987
Run 2	96.5	0.961	0.984
Run 3	97.0	0.963	0.986
Run 4	96.0	0.960	0.983
Run 5	97.5	0.966	0.985
Mean ± SD	96.8 ± 0.4	0.963 ± 0.003	0.985 ± 0.002

Key Confusion Matrix Terms for the Three Evaluated Models

Metric	LSTM	Isolation Forest	One-Class SVM
TP (True Positive)	97	93	90
FN (False Negative)	3	7	10
TN (True Negative)	396	388	384
FP (False Positive)	4	12	16

Simulation parameters

Parameter	Value
System	IEEE 118-bus
Library	Pandapower v2.13
Total sample	2000
Normal / FDIA	1600 / 400 (80% / 20%)
Test set	500 (400 normal + 100 FDIA)
Feature size	12 features per bus
Training / Test split	75% / 25%
Target bus	Bus 69

Recall Comparison by Attack Type

Attack Scenario	LSTM	Isolation Forest	One-Class SVM
Scenario A (Gradual Drift)	96.8%	89.4%	83.2%
Scenario B (Sudden Injection)	98.9%	97.1%	94.6%
Scenario C (Coordinated Corruption)	97.4%	91.3%	87.0%
Scenario D (Low Magnitude)	95.6%	85.7%	81.9%

5-Fold Walk-Forward Validation Results

Fold	LSTM Recall (%)	IF Recall (%)	OC-SVM Recall (%)
Fold 1	96.0	92.0	89.0
Fold 2	97.0	93.0	90.0
Fold 3	96.0	92.5	89.5
Fold 4	97.0	93.5	91.0
Fold 5	96.5	93.0	90.5
Mean ± SD	96.5 ± 0.6	92.8 ± 0.6	90.0 ± 0.7

Comparison of FDIA detection performances for a test set of 500 samples

Metric	LSTM	Isolation Forest	One-Class SVM
Accuracy	98.60%	96.20%	94.80%
	[97.14%–99.32%]	[94.14%–97.55%]	[92.49%–96.43%]
Recall	97.00%	93.00%	90.00%
	[91.55%–98.97%]	[86.25%–96.57%]	[82.56%–94.48%]
Precision	96.04%	88.57%	84.91%
	[90.26%–98.45%]	[81.08%–93.34%]	[76.88%–90.49%]
F1-Score	0.965	0.907	0.874
	[0.949–0.981]	[0.882–0.933]	[0.845–0.903]
AUC-ROC	0.987	0.965	0.942
	[0.971–1.000]	[0.940–0.990]	[0.910–0.974]
False Positive Rate	1.0%	3.0%	4.0%
Training Time (s)	42.3	0.8	4.2
95% Recall Benchmark	✓ PASSES	✗ FAILS	✗ FAILS

McNemar Test Results

Comparison	b	χ²	p	Significant?
LSTM vs Isolation Forest	12	10.083	0.0015	Yes**
LSTM vs One-Class SVM	19	17.053	<0.001	Yes***
Isolation Forest vs One-Class SVM	7	5.143	0.0233	Yes*

References

Authors

Metrics

Articles in this issue

DOI: https://doi.org/10.2478/ias-2026-0011 | Journal eISSN: 1554-1029 | Journal ISSN: 1554-1010

Journal RSS Feed

Language: English

Page range: 217 - 229

Published on: Jul 22, 2026

Published by: Cerebration Science Publishing Co., Limited

In partnership with: Paradigm Publishing Services

Publication frequency: 6 issues per year

Keywords:

FDIA,

LSTM,

Related subjects:

Fundamentals of computer sciences,

Theoretical computer sciences,

IT-security and cryptology

© 2026 Recep AKKAN, Osman Can ÇETLENBİK, published by Cerebration Science Publishing Co., Limited
This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 License.

Volume 21 (2026): Issue 4 (July 2026)

Detection of False Data Injection Attacks (FDIA) Targeting Distribution Transformers in Smart Grids Using Hybrid Artificial Intelligence Methods

Figures & Tables

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

LSTM Performance Stability Across 5 Independent Runs

Key Confusion Matrix Terms for the Three Evaluated Models

Simulation parameters

Recall Comparison by Attack Type

5-Fold Walk-Forward Validation Results

Comparison of FDIA detection performances for a test set of 500 samples

McNemar Test Results

Paradigm

My account