One of the main problems of a solar PV plant is low generation of electricity during bad weather condition when the generated power is less than the claimed demand of power. Under this condition it is not possible to generate more electricity as per demand once the power plant is designed. When the supply of electricity to the consumers reduces drastically and when there is no option for manipulation of power — a blackout out or load shedding is the inevitable. Online demand regulation i.e. regulating the claimed demand of the individual consumer depending on predicted power generation is an alternate option in this constrained situation. The present paper will address this particular problem through probabilistic duration estimation power against a critical load using statistical model. Our goal is to design a predictive model considering the environmental fluctuation of solar clarity index into consideration and incorporates a policy of meeting the critical base load primarily and loads exceeding the base load secondarily depending on predicted energy generation. A modified model of MARS (Multi variate adaptive regression spline) is developed and finally used for prediction of battery state of health from a remote end using sunlight intensity (similar to clarity index) using remote sensor technology. We use Multivariate adaptive smoothing spline with adaptive smoothing features of noisy data available from light sensor.Copyright © 2010 by ASME