A Hybrid Machine Learning Approach to Predicting Indirect Lightning Strike Impacts on Rooftop Solar PV Systems in Cagayan de Oro City, Philippines

v\:* {behavior:url(#default#VML);} o\:* {behavior:url(#default#VML);} w\:* {behavior:url(#default#VML);} .shape {behavior:url(#default#VML);} Normal 0 false false false false EN-GB X-NONE AR-SA /* Style Definitions */ table.MsoNormalTable {mso-style-name:"Table Normal"; mso-tstyle-rowband-size:0; mso-tstyle-colband-size:0; mso-style-noshow:yes; mso-style-priority:99; mso-style-parent:""; mso-padding-alt:0in 5.4pt 0in 5.4pt; mso-para-margin-top:0in; mso-para-margin-right:0in; mso-para-margin-bottom:8.0pt; mso-para-margin-left:0in; line-height:107%; mso-pagination:widow-orphan; font-size:11.0pt; font-family:"Aptos",sans-serif; mso-ascii-font-family:Aptos; mso-ascii-theme-font:minor-latin; mso-hansi-font-family:Aptos; mso-hansi-theme-font:minor-latin; mso-font-kerning:1.0pt; mso-ligatures:standardcontextual; mso-ansi-language:EN-GB;} Purpose: Solar photovoltaic systems installed on rooftops can be considered as advanced systems in the deployment of clean and renewable energy sources. However, these systems are considered to be vulnerable to the impacts of indirect lightning strikes, especially in areas that are commonly associated with storm activities, such as the Philippines. Indirect lightning strikes refer to those that do not strike the systems but can affect them in a negative manner. In the Philippines, the economic impacts of storm-related outages, including those associated with lightning strikes, can be very significant. It was noted that the economic cost of a 5-hour outage can be as high as P556 million. In this research, a smart model using machine learning was proposed with the aim of predicting the impacts of lightning strikes on solar photovoltaic systems. Design/methodology/approach: We have proposed a hybrid model that combines Conv1D, LSTM, attention, and XGBoost techniques to evaluate the severity of faults caused by a lightning strike. The proposed model has been provided with various environmental factors, characteristics of a lightning strike, and system specifications using actual dataset available from Cagayan de Oro City in the Philippines. v\:* {behavior:url(#default#VML);} o\:* {behavior:url(#default#VML);} w\:* {behavior:url(#default#VML);} .shape {behavior:url(#default#VML);} Normal 0 false false false false EN-GB X-NONE AR-SA /* Style Definitions */ table.MsoNormalTable {mso-style-name:"Table Normal"; mso-tstyle-rowband-size:0; mso-tstyle-colband-size:0; mso-style-noshow:yes; mso-style-priority:99; mso-style-parent:""; mso-padding-alt:0in 5.4pt 0in 5.4pt; mso-para-margin-top:0in; mso-para-margin-right:0in; mso-para-margin-bottom:8.0pt; mso-para-margin-left:0in; line-height:107%; mso-pagination:widow-orphan; font-size:11.0pt; font-family:"Aptos",sans-serif; mso-ascii-font-family:Aptos; mso-ascii-theme-font:minor-latin; mso-hansi-font-family:Aptos; mso-hansi-theme-font:minor-latin; mso-font-kerning:1.0pt; mso-ligatures:standardcontextual; mso-ansi-language:EN-GB;} The proposed model has been compared with other simple models such as Random Forest, SVM, and a simple LSTM model using metrics such as accuracy, F1 score, and area under the receiver operating characteristic curve (AUC) with 5-fold cross-validation. Findings: The model recorded an accuracy of 91.2%, and the average AUC was 0.93, which was higher than the baseline classifiers. The use of attention heat maps and feature importance helped identify the dominance of lightning peak current, lateral distance, and EMI intensity in the model. The model was highly generalizable with class imbalance and was interpretable through visual methods. The model can be used for deployment in SCADA systems and edge devices for real-time risk prediction. Originality: This work is the first to combine temporal deep learning with physical system models to predict indirect effects of lightning strikes on photovoltaic systems. This new model maintains long-range temporal dependencies and spatial surge patterns while keeping the model interpretable. This model closes the gap between static protection schemes and dynamic risk management.