A method for short term load forecasting in a power grid includes obtaining historical data comprising power data, load data and weather data corresponding to time index data recorded from a location in a power distribution network of the power grid. The method further includes receiving power grid data comprising a plurality of power values, and a plurality of weather parameter values corresponding to a plurality of recent time instant values. The method also includes generating modified historical data using statistical techniques to rectify error conditions. The method further includes estimating one or more power values at a future time instant based on the modified historical data and the power grid data. The method also includes balancing load of the power distribution network based on the estimated one or more power values.