In modern photovoltaic (PV) power generation systems, the efficient operation of solar panels is critical for maximizing energy output and ensuring system reliability. However, the large-scale deployment of solar panels in PV plants makes it challenging to manually inspect each panel for anomalies, such as power degradation, voltage instability, or current irregularities. Traditional methods often struggle with accuracy and speed due to environmental factors, equipment aging, and operational variances. To address these limitations, we propose a novel anomaly detection approach for solar panels that leverages the concept of relative power generation feature matching. This method focuses on analyzing the alignment between actual and expected power generation characteristics of solar panels, enabling rapid identification of abnormal behaviors. By modeling abnormal power actions and deriving key parameters like anomalous voltage and current values, we establish a comprehensive detection framework. Our experiments demonstrate that this approach effectively identifies faulty solar panels, enhancing the stability and efficiency of PV systems. Throughout this paper, we emphasize the importance of monitoring solar panels to maintain optimal performance.
The core of our method lies in defining the relative power generation feature matching degree, which quantifies the similarity between the actual power generation signals of solar panels and their expected characteristics. Let α represent the feature parameter of the power generation signal, δ denote the expected power generation characteristic parameter, mα and mδ be the power level coefficients based on α and δ, respectively, and b be a matching operation coefficient. The relative power generation feature matching degree M is expressed as:
$$ M = \frac{(m_{\alpha} \cdot m_{\delta})^2}{b (\alpha – \delta)^2} $$
This formula captures the degree of alignment in power generation features for solar panels. To detect anomalies, we set an anomaly threshold v0 and compare it with the relative difference v′ between actual and expected power outputs. When v′ > v0, we sample abnormal power behavior using the following condition, where c1 and c2 are actual and expected power valuation terms, respectively:
$$ B = \min \left| M \cdot \left| \frac{v’}{c_1} \right|^2 – \left| \frac{v_0}{c_2} \right|^2 \right| $$
Next, we model the abnormal power behavior by considering a set of random power parameters X = {x1, x2, …, xn} sampled from non-normal intervals, as defined by:
$$ x_1, x_2, \cdots, x_n \in (-\infty, 0) \cup (0, +\infty) $$
$$ x_1 \neq x_2 \neq \cdots \neq x_n $$
Combining these, the abnormal power behavior result Z, which accounts for the relative matching degree, is derived as:
$$ Z = \lg B \cdot (\beta – 1) \cdot \frac{l}{X} $$
Here, β is a non-normal power definition parameter, and l is the anomaly level coefficient for solar panels based on relative matching. This modeling allows us to swiftly capture deviations in the power generation of solar panels, facilitating early anomaly detection.
To further refine the anomaly detection for solar panels, we calculate the anomalous voltage and current values, as power anomalies often stem from voltage and current irregularities. In PV systems, solar panels are typically connected in series, leading to additive anomalous voltage values. Let E represent the electrical signal load characteristic parameter, χ be the power host performance definition parameter, q denote the charge of the power signal, ε be the inductance coefficient, f represent the charge load parameter under anomalous voltage, and W0 be the standard power generation per unit time. The anomalous voltage value U is computed as:
$$ U = \frac{E}{\chi^2} \cdot Z \cdot (f \cdot q \cdot \varepsilon – W_0) $$
In contrast, the current in series-connected solar panels remains consistent, so anomalous current values do not accumulate. Defining ΔP as the unit output of the power generation signal, φ as the point charge sampling coefficient, ϕ as the charge load parameter under anomalous current, and h as the transmission current load characteristic parameter, the anomalous current value I is given by:
$$ I = Z^{-1} \cdot \left( \frac{E}{W_0} \right) \cdot \varphi \cdot |\Delta P|^2 + \frac{1}{\phi \cdot h} $$
These calculations enable us to establish a unified detection criterion for anomalies in solar panels. By considering the operational phases of solar panels, we define the detection result K using constraint coefficients and thresholds. Let λ be the abnormal power signal constraint coefficient, k represent the anomalous operation characteristic of solar panels, d be the standard detection threshold, ι denote the constraint term for power generation behavior based on relative matching, and g represent the abnormal power signal. The anomaly detection result is expressed as:
$$ K = \frac{2 \lambda (k – 1)}{d – \iota – g} \cdot \left( \frac{I}{U} \right)^3 $$
This formula integrates voltage and current anomalies to provide a comprehensive detection standard for solar panels, reducing false positives and missed detections. The relative power generation feature matching enhances the sensitivity to deviations, ensuring that anomalies in solar panels are identified promptly.
To validate our approach, we conducted experiments under different anomaly scenarios, such as complete damage and surface scratches on solar panels. The experimental setup involved connecting 10 identical solar panels in series, each with a maximum power of 80 W, and monitoring the total power output in the PV circuit. We compared our method with existing techniques, including multidimensional time series and VMD-XGBoost-BiLSTM models, focusing on the accuracy of detecting abnormal power behaviors in solar panels. The key parameters for the experiments are summarized in the table below.
| Parameter Name | Value |
|---|---|
| Rated Voltage (V) | 220 |
| Rated Current (A) | 45 |
| Rated Power (W) | 800 |
| Number of Series-Connected Solar Panels | 10 |
| Maximum Power per Solar Panel (W) | 80 |
In the damage anomaly experiment, where solar panels were completely non-functional, our method showed a significant drop in power output, reaching zero at the 50-minute mark, with an overall lower mean power compared to other methods. This indicates a high sensitivity to severe anomalies in solar panels. For scratch anomalies, which cause partial degradation, our approach consistently detected lower power values, with a minimum of 400 W by the end of the experiment, outperforming the compared methods in identifying subtle irregularities. The results are summarized in the following table, highlighting the average power outputs and detection accuracy for solar panels under different anomaly conditions.
| Anomaly Type | Our Method | Multidimensional Time Series | VMD-XGBoost-BiLSTM Model |
|---|---|---|---|
| Damage | 120 | 450 | 460 |
| Scratch | 420 | 580 | 590 |
The superior performance of our method can be attributed to the incorporation of relative power generation feature matching, which allows for dynamic adjustment to the operational states of solar panels. By continuously monitoring the matching degree, our system can detect both abrupt and gradual anomalies in solar panels, such as those caused by environmental stress or physical damage. This proactive detection helps in maintaining the overall health of PV systems, ensuring that solar panels operate at peak efficiency. Furthermore, the use of derived voltage and current values provides a multi-faceted view of anomalies, reducing the reliance on single-parameter assessments that may miss complex fault patterns in solar panels.
In conclusion, our anomaly detection method for solar panels, based on relative power generation feature matching, offers a robust solution for identifying faults in large-scale PV systems. By modeling abnormal power behaviors and calculating anomalous voltage and current values, we achieve high accuracy and reliability in detecting various types of anomalies in solar panels. The experimental results confirm that this approach can swiftly identify degraded solar panels, contributing to improved system stability and energy output. Future work could explore the integration of machine learning techniques to further enhance the adaptability of this method to diverse environmental conditions affecting solar panels.