Power Quality Detection Using LSTM Deep Learning (AI)

Learn power quality assessment by extracting features from voltage and current signals using Fourier transform, S-transform, the wavelet transform, and Hilbert-Huang transforms to classify disturbances like sags and harmonics.

Discover automatic power quality classifiers that pre-process waveforms, segment and extract features, classify disturbances, and decide whether a transient, flicker, voltage sag, or harmonic distortion occurred.

Bundle preprocessing steps to prepare the waveform for power quality classification: segment and normalize signals, identify event segments between transitions, and extract robust features from stationary event windows.

The Fourier transform converts time-domain signals to the frequency domain, revealing the frequency spectrum and aiding power quality analysis, while highlighting limitations for short-term transients.

Learn how the discrete Fourier transform analyzes finite signals by converting time to frequency with FFT, using windowing, and noting time information loss. Explore STFT and wavelet transforms.

Explore the fast Fourier transform (FFT) for efficient discrete Fourier transform calculations, enabling frequency estimation and harmonic analysis with windowing, and discuss combining with DWT to detect power quality events.

Use the short-time Fourier transform (STFT) to reveal time–frequency content of power signals with a sliding window. Learn the time–frequency trade-off in detecting and classifying power quality problems.

Apply the discrete short-time Fourier transform to analyze non-stationary power signals by windowing segments, computing fast Fourier transforms, and tracking amplitude and phase over time for faults, transients, and harmonics.

Investigate how the continuous wavelet transform enables time–frequency analysis for power quality, detecting voltage sags, transients, and harmonics with multi-resolution insight.

Explore how discrete wavelet transform uses dyadic filters and downsampling for multi-resolution analysis. Identify transitions with time-frequency details through approximation and detail coefficients for power quality and fault detection.

Discover the wavelet packet transform (WPT) as a flexible extension of the DWT that decomposes both approximation and detail to yield a richer time-frequency view of non-stationary power signals.

Explore the Stockwell transform (S-transform) as a time-frequency localization method, hybridizing STFT and CWT with a frequency-dependent Gaussian window that preserves phase and tracks frequency content for power quality analysis.

Explore the discrete S-transform, a time-frequency tool that blends Fourier and wavelet ideas for efficient signal analysis. Learn its Gaussian window localization, FFT-based acceleration, and O(n^2) complexity.

Apply the Hilbert–Huang transform to decompose non-stationary signals into intrinsic mode functions by empirical mode decomposition and analyze their frequency, amplitude, and phase, enabling envelope tracking in power engineering.

Explain how the discrete Gabor transform uses a Gaussian window to localize signals in time with improved time-frequency resolution over the short-time Fourier transform, yielding smoother representations.

Explore how the Kalman filter uses state-space equations to predict and correct estimates of hidden state vectors from noisy measurements, enabling real-time tracking in power systems.

Compare Fourier, wavelet, and S-transform with STFT, HHT, and Gabor transforms, noting that the S-transform bridges Fourier and wavelet and that redundancy and sampling trade-offs for stationary versus non-stationary signals.

Identify and classify power quality disturbances automatically using AI-driven classifiers such as artificial neural networks (ANN), support vector machines (SVM), and fuzzy expert systems (FES).

Explore artificial neural networks for power quality classification, using voltage, current, and frequency inputs to predict labels such as voltage sag or harmonic distortion, including probabilistic neural networks.

Explore how support vector machines handle large, multi-class classification with kernel functions, 1v1 and 1v rest strategies, and quadratic programming for nonlinear separation and strong generalization in power quality analysis.

We explore the expert system (ES) and fuzzy logic, showing how degrees of truth and if-then rules help diagnose power quality events, while noting limitations in scalability and portability.

Compare the artificial neural network, support vector machine, and expert system in power quality classification, outlining their advantages, disadvantages, and when to prefer each approach.

Translate the classifier's output in the decision-making stage into a power-quality event type—such as voltage sag, transient, or harmonic distortion—aided by expert systems or fuzzy logic to boost accuracy.

Introduce an automatic online power quality detection system that classifies disturbances from real-time voltage and current waveforms using LSTM-based learning with automatic training data and no manual feature extraction.

Apply an LSTM-based system to automatically detect and classify power quality events using amplitude, start and end times, duration, and THD, producing waveform class and harmonics indication.

Explore how recurrent neural networks remember past inputs to analyze time-dependent data, including speech, video, and power system waveforms. See how the vanishing gradient led to LSTM for long-range memory.

Explore how LSTM networks overcome long-term dependencies in time series by using memory cells and gates to manage information flow, enabling accurate predictions in power quality monitoring.

Develop real-time power quality detection using an lstm to classify waveforms into interruption, sag, flicker, normal, swell, or surge, and to indicate harmonics with a second output.

Generate and test power quality disturbances using MATLAB simulations to evaluate an LSTM model, analyzing sag, swell, interruption, flicker, surge, and harmonics with varying SNR.

Identify the normal voltage waveform as the base case: a pure, stable waveform with no rms variation, representing the ideal, perfectly normal condition.

Demonstrates a noise-free normal voltage waveform with negligible THD, validated by LSTM outputs. Shows predicted magnitudes closely match actual values, achieving over 99.4 percent accuracy.

Demonstrates that a normal voltage waveform with SNR 40 dB remains clean despite noise, with LSTM outputs showing no distortion and classification accuracy near 99.4%.

Explore power quality detection using an LSTM to classify a normal voltage waveform under SNR 30 dB, with THD under 0.57% and prediction accuracy of 98.96%.

Explore how a LSTM deep learning model detects a normal voltage waveform under SNR 20 dB, showing low THD, no distortion, and nearly 98% accuracy in classification and prediction.

Detect voltage interruptions with a noise-free waveform using an LSTM model, achieving nearly 100% accuracy in classifying interruptions and predicting amplitudes, durations, and THD.

Analyze how a voltage interruption waveform with SNR 40 dB behaves under noise and how an LSTM detects interruption, absence of harmonics, and predicts magnitude, duration, and THD.

Demonstrates an LSTM model detecting voltage interruption under SNR 30 decibels, confirming no harmonic distortion, and accurately predicting magnitude, duration, and THD with high accuracy.

Assess voltage interruption waveforms at 20 dB SNR, showing LSTM detects interruptions, confirms no harmonic distortion, and predicts magnitude, timing, and THD with accuracy above 90%.

Define voltage sag as a short-duration rms voltage reduction caused by sudden load increases or short circuits, in the context of power quality detection using lstm deep learning.

Identify voltage sag waveform without noise using an LSTM model that outputs 0.5 for sag and 0 for no harmonic distortion, achieving 99.96% accuracy and THD below 2%.

Demonstrates voltage-sag detection using LSTM deep learning at SNR 40 dB, with outputs signaling sag presence and no harmonic distortion, and achieves 97.51% accuracy.

Apply LSTM deep learning to identify voltage-sag waveforms at 30 dB SNR, achieving 95.52% accuracy with low THD and no harmonic distortion.

Detects a voltage sag waveform at an SNR of 20 dB using LSTM deep learning, achieving over 92% accuracy and THD under 1.7%, with no harmonic distortion.

LSTM identifies distorted and noisy voltage waveforms without added noise, reveals harmonic content with THD above 18%, and achieves 98.9% prediction accuracy.

Analyze voltage distortion waveforms with a 40 dB SNR and see how an LSTM detects distortion and noise with over 97.5% accuracy in predicting per-unit magnitudes and THD above 18.6%.

Demonstrates that an LSTM model detects voltage distortion and noise at 30 dB SNR, with THD above 18.4% and about 96% accuracy in predicting THD and waveform distortion.

The lecture demonstrates detecting a distorted, noisy voltage waveform at an SNR of 20 dB using an LSTM, with THD above 18.5% and about 94.6% accuracy.

Explore voltage swell, a brief overvoltage where the rms voltage rises above normal for a short period in the waveform before returning to normal.

Identify a clean voltage swell using an LSTM model, noting swell magnitudes from 1.1 to 1.8 per unit with 0.15 seconds duration, THD below 1%, and classification accuracy around 99.83%.

Demonstrating a voltage-swell waveform with SNR 40 dB, the lecture shows low THD and robust LSTM predictions, achieving 99.2% accuracy with no harmonic distortion.

Voltage swell waveform with SNR 30 dB is detected by an LSTM model under noisier conditions, achieving about 96.87 percent accuracy and confirming no harmonic distortion.

Evaluate how an LSTM model detects voltage swell at an SNR of 20 dB, remains accurate under noise, and reports low THD and no harmonic distortion.

Detect and classify voltage surges with LSTM deep learning, quantify harmonic distortion with THD, and accurately predict surge magnitude and timing in noise-free waveforms.

Detect a voltage surge with a 40 dB SNR, where the lstm identifies the surge with no harmonic distortion and 99.77% prediction accuracy, THD below 0.36%.

Explore voltage surge waveforms at an SNR of 30 dB and evaluate how noise impacts LSTM classification and magnitude prediction with low harmonic distortion.

The lecture demonstrates LSTM-based detection of voltage surges with an SNR of 20 dB, showing low THD, accurate magnitude predictions, and no harmonic distortion.

Assess voltage flicker as a noise-free, slow RMS voltage variation. LSTM predicts magnitudes close to actual values with low THD and about 99% accuracy, confirming no harmonic distortion.

Demonstrates robust voltage flicker detection with a 40 dB SNR, no harmonic distortion, and an LSTM model, achieving about 98% accuracy and THD below 0.28% under noise.

Explore how an LSTM deep learning model detects voltage flicker waveform in a noisy signal with SNR 30 dB, achieving about 93% accuracy and low THD under 0.59%.

Explore how an LSTM detects a voltage flicker waveform at SNR 20 dB, shows no harmonic distortion, and maintains about 92% accuracy despite noise.

Detect voltage sag with harmonics using lstm deep learning, showing precise identification of sag and harmonic distortion in clean waveforms with high thd.

Analyze voltage sag with harmonics under SNR 40 dB, using LSTM outputs to detect distortion and predict THD, achieving over 98% accuracy.

Examine voltage sag with harmonics at snr 30 db, using lstm outputs to identify sag and distortion with up to 98.53% classification accuracy and 97.2% prediction accuracy.

Detect voltage sags with harmonics at a 20 dB SNR using LSTM deep learning, achieving about 98.8% overall accuracy in identifying sag, duration, and THD.

The LSTM model accurately detects voltage swell with harmonics without noise, predicting magnitude, duration, and THD with about 99.8% accuracy, demonstrating robust classification of harmonic distortion in power quality.

Detect voltage swell with harmonic distortion and noise at SNR 40 dB using an LSTM, achieving over 97.9% prediction accuracy and 99.84% overall classification.

Demonstrates how an LSTM model detects a noisy voltage swell with harmonics at 30 dB SNR, predicting magnitude, duration, and THD with high accuracy.

Assess the LSTM model’s ability to identify a noisy voltage swell with harmonics at 20 dB SNR. It delivers high accuracy and robustly predicts magnitude, duration, and THD.

Compare the proposed LSTM technique with established methods for power quality classification, highlighting 100% noiseless accuracy and strong robustness across 20–40 dB noise.

Explore real-world power quality detection with an LSTM on a 66-11 kV substation, using real three-phase voltages and currents captured by a high-rate data acquisition system in MATLAB.

Analyze practical three-phase voltage data ul1, ul2, and ul3 using LSTM to detect and classify power quality events such as voltage swell, voltage sag, or normal conditions.

Explore how an lstm deep learning model detects and classifies power quality events in three-phase currents i1, i2, and i3, including swell with harmonics and harmonic distortion.

Analyze three-phase voltages UL1, UL2, and UL3 with an LSTM model to detect and classify power quality events, including swells and sags without harmonics.

Analyze simultaneous 3-phase voltages and currents with LSTM to detect normal and sag events across UL1, UL2, and UL3, highlighting THD and power quality classifications.

Analyze three-phase currents I1, I2, I3 with LSTM to identify power-quality events—swells and sags—without harmonics.

Assess three-phase voltages UL1, UL2, UL3 and current I1 with an LSTM model to classify power quality, finding UL1 and UL2 as interruptions with harmonics, UL3 normal, and I1 distorted.

Demonstrate that an lstm detects and classifies power quality events using amplitude, start time, end time, duration, thd%, and outputs waveform class and harmonics indication, validated on simulations and data.