Individual Baseline Modeling allows flexibility in the way Baseline Models are defined.

The Coefficient of Variation of the Root Mean Squared Error, or CV(RMSE), is one measure for quantifying model quality. This measures the differences between values predicted by a model and the values actually observed. A lower value indicates less variance and hence higher quality. When selecting a model give preference to those exhibiting low CV(RMSE)

T-Stat values are also used to determine the validity of the model. The higher the value, the more valid the model. Normally a valid model would have a T-Stat values for all independent variables over 2.0.

If we use Q to represent the expected usage in whatever physical unit is being used, the example above can be read as:

Q = (number of days in the billing period x 2149) + (Heating Degree Days in the billing period at a balance point of 18C x 46.3)

Your portfolio manager will most likely have defined a meter modeling standard that you should use to ensure a consistent approach for all meters within a project. |

Typical Utility Meter Modeling Standard |

1. Select meter readings from a period representing all of the operating modes of the facility. Typically this will be a full year containing all four seasons. |

2. Set balance temperatures for heating and cooling at 18°C. |

3. Select HDD and CDD as independent variables based on judgment and referring to on-screen graphs. |

4. Optimize balance temperatures accordingly to improve CV(RMSE) and T-Statistics. |

5. Remove outlying data points by deselecting them in the billing data selector. This should only be done if you can justify why this data point represents abnormal utility usage. |

6. Ideally, the magnitude of the T-Statistic for each variable should be greater than 2.0. Variables exhibiting T-Statistic outside of this recommendation can be included but their impact on utility usage is negligible. |

7. Ideally, the CV(RMSE) value should be less than or equal to 5% as per IPMVP and ASHRAE Guideline 14. However in practice this is not always possible unless all significant utility use variables have been identified and represented in the model. |

S5_Meter_Modeling_Standard ©2015 Managing Energy Inc.