Evaluating Cyber-Physical Power Converter Systems Using MATLAB/Simulink

In the rapidly evolving field of power electronics, the integration of cyber-physical systems (CPS) is becoming increasingly vital. This article delves into a proposed strategy for evaluating a cyber-physical power converter-based system using the MATLAB/Simulink environment. The system comprises various components, including a DC source, a DC/DC buck power converter, and an ohmic load. We will explore the system’s configuration, the implementation of an artificial neural network (ANN) for false data detection, and the results of various scenarios tested within this framework.

System Configuration

The cyber-physical power converter system under consideration consists of a DC source ((V1)), a buck DC/DC power converter, and an ohmic load ((R{Out})). The DC source is set at (V1 = 145V), while the load resistance is (R{Out} = 1.9 \Omega). The output voltage of the power converter is controlled to a reference value of (R_v = 50V). A crucial aspect of this study is the introduction of a time delay ((\sigma = 1)) in the system, which simulates real-world communication delays that can occur in cyber-physical systems.

To ensure practical evaluation, a sampling time ((\Delta t)) of (0.1ms) is selected, aligning with previous experimental validations in power electronics that utilized sampling frequencies of 10 kHz or higher. The implemented ANN consists of an input layer with 6 neurons, a hidden layer with 10 neurons, and an output layer with a single neuron to estimate the value of false data.

Training the ANN

The ANN is trained using a dataset generated from the operation of the DC/DC converter over a period of 3 seconds. Two distinct modes are employed: normal operation without false data and operation under false data conditions. The training dataset comprises 30,000 samples, and the Levenberg-Marquardt algorithm is utilized for training, with performance evaluated using a mean squared error (MSE) strategy.

The dataset is divided into three groups: 70% for training, 15% for testing, and 15% for validation. The regression plots for training, validation, and testing are depicted in the results, showcasing the ANN’s ability to learn and adapt to the data.

Scenario Analysis

The study evaluates the proposed strategy under six different scenarios, each designed to test the system’s resilience against various types of false data injections.

Scenario 1: Constant False Data

In this scenario, the power converter operates normally until (t = 1.5s), when constant false data values (1V, 3V, 4V, 5V, and 5.5V) are injected into the system. The output voltage of the DC/DC converter is monitored, revealing a significant deviation from the reference voltage of 50V. The ANN successfully estimates the injected false data, demonstrating its effectiveness in identifying constant false data inputs.

Scenario 2: Sine-Based Time-Varying False Data

Here, a sine function with an amplitude of 4.8V is introduced at (t = 1s). The output voltage begins to fluctuate in accordance with the sine wave, indicating the presence of false data. The ANN accurately tracks the sine-based false data, showcasing its capability to adapt to time-varying inputs.

Scenario 3: Sawtooth-Based Time-Varying False Data

In this scenario, a sawtooth wave with a frequency of 1Hz and an amplitude of 4V is injected at (t = 1s). The output voltage reflects the sawtooth pattern, and the ANN effectively estimates the false data, further validating its robustness against different waveforms.

Scenario 4: Gaussian-Based Time-Varying False Data

A Gaussian-based false data signal is introduced at (t = 1s). The output voltage fails to track the reference voltage, indicating the impact of the false data. The ANN demonstrates its ability to monitor and estimate the Gaussian-based false data accurately.

Scenario 5: Symmetrical Communication Delays and False Data

In this scenario, a false data value of 6V is injected at (t = 1s) while symmetrical communication delays of 1ms are applied to the ANN inputs. The ANN exhibits a transient response, requiring a brief period to adjust to the communication delay before accurately estimating the false data.

Scenario 6: Unsymmetrical Communication Delays and False Data

Finally, unsymmetrical communication delays (1ms, 2ms, and 3ms) are introduced for different inputs. The ANN continues to estimate the false data effectively, albeit with a transient period required for adjustment.

Conclusion

The evaluation of the cyber-physical power converter-based system using MATLAB/Simulink demonstrates the effectiveness of the proposed strategy in detecting and estimating false data under various scenarios. The ANN’s ability to adapt to different types of false data, including constant, time-varying, and Gaussian-based inputs, highlights its potential for enhancing the cybersecurity of power electronics systems.

As cyber-physical systems become increasingly prevalent, the integration of advanced monitoring and detection strategies will be essential for ensuring their reliability and security. The findings of this study pave the way for further research into robust control strategies and the development of resilient power converter systems in the face of cyber threats.