Fine　description　for　regional　solar　spectra　has　always　been　critical　for　improving　solar　energy　utilization.　However,　it　is　difficult　and　complex　to　implement　due　to　large　amount　and　high-dimensional　characteristics　of　the　measured　solar　spectra.　To　address　the　above　challenges,　this　study　proposed　a　data-driven　method　to　describe　regional　solar　spectra　using　211,877　global　horizontal　irradiance　spectra　in　Beijing　area　as　case　data.　Firstly,　the　dimensions　of　each　measured　spectrum　were　reduced　from　2,221　to　six　using　the　deep　autoencoder.　Then,　all　dimensionality-reduced　measured　spectra　were　categorized　into　five　clusters　using　the　agglomerative　hierarchical　clustering.　The　clustering　results　exhibited　notable　variations　across　different　months　and　dates.　Finally,　the　local　reference　spectra　were　determined　based　on　clustering　results,　and　their　effects　on　the　performance　of　typical　photovoltaic　materials　were　analyzed.　The　maximum　mismatch　factor　for　typical　photovoltaic　materials　under　local　reference　spectra　can　reach　up　to　21　%.　The　comparisons　with　standard　spectrum　highlighted　that　the　superior　capacity　of　local　reference　spectra　to　describe　the　configuration　of　regional　solar　energy.　This　study　provides　a　novel　insight　into　the　fine　description　for　regional　solar　spectra,　which　sets　a　new　direction　for　innovation　in　photovoltaic　technology　and　promotes　the　sustainable　utilization　of　renewable　energy.