<p>Hot spots are common defects in photovoltaic (PV) modules that can lead to performance degradation and even pose a fire hazard. This study proposes an online detection methodology for hot spots within a data-driven framework. The modeling of PV modules under normal conditions relies on the integration of two-dimensional information. The first dimension pertains to the correlation among measured variables across multiple tests, while the second dimension captures the local nonlinearity among past, present, and future measurements from PV modules. By combining these two dimensions of information, accurate modeling of PV modules can be achieved in a data-driven manner. Based on this model, the performance-oriented detection of hot spots in PV modules is formulated in detail. The proposed approach is then applied to both darkroom and outdoor PV systems, with a total of 23 tests and comparisons conducted. The results of these tests and comparisons convincingly demonstrate the validity and superiority of the proposed method.</p>