Method and system for forecasting building power consumption

A system for forecasting a power consumption of one or more buildings for a predefined forecast horizon comprising:

• an input/output interface to receive a historical power consumption data and corresponding range of temperature values, an occupancy data, and a location of the one or more buildings from one or more historical databases, wherein the one or more historical databases maintains physical sensors data of power consumption along with the sensory information useful in the one or more power consumption including occupancy and weather information, wherein the occupancy and weather information is pre-processed to remove discrepancies prior to entering into the historical databases;
• one or more hardware processors;
• a memory in communication with the one or more hardware processors, wherein the one or more hardware processors are configured to execute programmed instructions stored in the memory, to:
• pre-process the received historical power consumption data to remove one or more redundant data, one or more outliers, one or more unreal values from the received historical power consumption data and impute missing values of the received historical power consumption data;
• synchronization of smart meter data with sensory information received from data sources including occupancy and weather information;
• re-arrange the received historical power consumption data to bring proximity between varied time instants of energy consumption by fixing a window on the historical power consumption data, wherein the received historical power consumption data corresponds to a time-of-the-day, a day-of-the-week, a season, and the location of the one or more buildings;
• calculate an energy-temperature correlation using the pre-processed historical power consumption data and a received historical range of temperature value;
• estimate an energy drift using the pre-processed historical power consumption data, wherein energy drift is a change in an average power consumption of a predefined month in consecutive years;
• select at least one model based on the estimated energy drift and the calculated energy-temperature correlation, wherein the at least one model includes one or more of a seasonal support vector regression (SSVR) model, a support vector regression (SVR) model, or an artificial neural network (ANN) model;
• derive one or more auto-regressive (AR) features from the pre-processed historical power consumption data using a partial auto-correlation function;
• generate a predictor matrix using the one or more derived AR features, and the received historical information of occupancy, temperature, and holidays of the one or more buildings;
• train the at least one selected model using the generated predictor matrix;
• forecast the power consumption of the one or more buildings for the predefined forecast horizon using the trained at least one model and a real-time information of occupancy, temperature, weekdays, holidays, and the location of the one or more buildings;
• and
• measure by a smart meter, the total power consumption of the one or more buildings in predefined minutes.

A non-transitory computer readable medium storing one or more instructions which when executed by one or more processors on a system, cause the one or more processors to perform method for forecasting a power consumption of one or more buildings for a predefined forecast horizon comprising:

• receiving, via an input/output interface, a historical power consumption data and corresponding range of temperature values, an occupancy data, and a location of the one or more buildings from one or more historical databases, wherein the one or more historical databases maintains physical sensors data of power consumption along with the sensory information useful in the one or more power consumption including occupancy and weather information, wherein the occupancy and weather information is pre-processed to remove discrepancies prior to entering into the historical databases;
• pre-processing, via one or more hardware processors, the received historical power consumption data to:
• remove one or more redundant data, one or more outliers, one or more unreal values from the received historical power consumption data;
• and
• impute missing values of the received historical power consumption data;
• synchronization of smart meter data with sensory information received from data sources including occupancy and weather information;
• re-arrange the received historical power consumption data to bring proximity between varied time instants of energy consumption by fixing a window on the historical power consumption data, wherein the received historical power consumption data corresponds to a time-of-the-day, a day-of-the-week, season, the location of the one or more buildings;
• calculating, via the one or more hardware processors, an energy-temperature correlation using the pre-processed historical power consumption data and a received historical range of temperature value;
• estimating, via the one or more hardware processors, an energy drift using the pre-processed historical power consumption data, wherein energy drift is a change in an average power consumption of a predefined month in consecutive years;
• selecting, via the one or more hardware processors, at least one model based on the estimated energy drift and the calculated energy-temperature correlation, wherein the at least one model includes one or more of a seasonal support vector regression (SSVR) model, a support vector regression (SVR) model, or an artificial neural network (ANN) model;
• deriving, via the one or more hardware processors, one or more auto-regressive (AR) features from the pre-processed historical power consumption data using a partial auto-correlation function;
• generating, via the one or more hardware processors, a predictor matrix using the one or more derived AR features, and the received historical information of occupancy, temperature, and holidays of the one or more buildings;
• training, via the one or more hardware processors, the at least one selected model using the generated predictor matrix;
• forecasting, via the one or more hardware processors, the power consumption of the one or more buildings for the predefined forecast horizon using the trained at least one model and a real-time information of occupancy, temperature, weekdays, holidays, and the location of the one or more buildings;
• and
• measuring by a smart meter, a total power consumption of the one or more buildings in predefined minutes.