Costs of PV installations has been progressively declining over the past decade and has attained grid parity in several countries. In developing countries such as India with high energy shortages and growing energy demand, new energy firms are setting up large solar farms to offering greener energy at a competitive tariffs. The fact that solar PV power is irradiance dependent make it an intermittent source of energy and thus prone to penalties while feeding the power into the national grid. Moreover, there are factors such as accumulation of dust, breakage, aging, etc. which can lead to under-performance of the solar farms, thus further reducing the profit margins. This calls for methodologies that can not only predict the power generation capacity of the solar farm accurately but also detect the under-performance, malfunction and aging of the solar panels. In this paper, we propose a moving window based machine learning approach to accurately predict the energy production from a solar farm. The proposed methodology may also be used monitor the performance yields of solar farms, identification of operational problems and degradations in solar panels due to aging. We demonstrate the efficacy of the proposed approach on a 1.2MW solar farm located in a tropical country consisting of 6 different PV technologies of 200KW each.