.Buildings.Controls.Predictors.ElectricalLoad

Information

Data-driven model that predicts the electrical load. This load prediction can for example be used in a demand response client.

The model computes either an average baseline or a linear regression with respect to outside temperature. For both, optionally a day-of adjustment can be made.

Computation of baseline

Separate loads are computed for any types of days. The type of day is an input signal received from the connector typeOfDay, and must be equal to any value defined in Buildings.Controls.Types.Day. This input is a vector where the first element corresponds to the current day, the next element to tomorrow, and so on. The dimension of this input vector is typically 2 if the demand is to be predicted for the next 24 hours. If it is for the next 48 hours, then the dimension is 3. Using a vector is required as the prediction could be from noon of a workday to noon of a holiday or week-end day.

The average baseline is the average of the consumed power of the previous n_his days for the same time interval. The default value is n_his=10 days. For example, if the prediction is mode for 1 hour time windows, then there are 24 baseline values for each day, each being the average power consumed in the past 10 days that have the same typeOfDay.

The linear regression model computes the predicted power as a linear function of the current outside temperature. The two coefficients for the linear function are obtained using a regression of the past n_his days.

If the input signal storeHistory is true, then the prediction is no longer carried out for this day until midnight. For example, if used for a demand respond client, on an event day, one may want to set storeHistory=true when the building operates in demand respond mode. Then any time interval after this signal is received is excluded from the baseline computation. Storing history terms for the baseline resumes automatically at midnight.

If no history term is present for the current time interval and the current type of day, then the predicted power consumption PPre[:] will be zero.

Day-of adjustment

If the parameter use_dayOfAdj = true, then the day-of adjustment is computed. (Some literature call this morning-of adjustment, but we call it day-of adjustment because the adjustment can also be in the afternoon if the peak is in the late afternoon hours.) The day-of adjustment can be used with any of the above baseline computations. The parameters dayOfAdj_start and dayOfAdj_end determine the time window during which the day-of adjustment is computed. Both need to be negative times, measured in seconds prior to the time at which the power consumption is predicted. For example, to use a day-of adjustment for the window of 4 to 1 hours prior to the event time, set dayOfAdj_start=-4*3600 and dayOfAdj_end=-3600.

The day-of adjustment is computed as follows: First, the average power P_ave consumed over the day-of time window is computed. Next, the average power P_his is computed for the past n_his days. Then, the adjustment factor is computed as

a = min(a_max, max(a_min, P_ave ⁄ P_his)),

where a_min and a_max are the minimum and maximum adjustment factors as defined by the parameters adjFacMin and adjFacMax.

Revisions

• June 9, 2015 by Michael Wetter:
  Updated model to allow translation using pedantic mode in Dymola 2016. This is for #426.
• October 29, 2014, by Michael Wetter:
  Revised implementation.
• October 25, 2014, by Michael Wetter:
  Implemented new assignment of the start of the sampling.
• September 6, 2014 by Michael Wetter:
  Moved to package Buildings.Controls.Predictors.
• July 10, 2014 by Michael Wetter:
  First implementation.
• March 20, 2014 by Michael Wetter:
  First implementation.