Unresonable load change when modifying the model inputs

edit

I wanted to modify the input of the window to see the change of energy consumption. And I used the secondary school of the DOE prototype building (ASHRAE90.1-2007). The weather file is Rochester, NY (6A). In the DOE prototype building, the initial setting related to window of secondary school is:

WindowMaterial:SimpleGlazingSystem,
  Nonres Window Glazing Layer,           !- Name
  2.7255648,              !- U-Factor {W/m2-K}
  0.4,                     !- Solar Heat Gain Coefficient
  0.4;                     !- Visible Transmittance

I tried to modify the input via reducing the U factor and increasing the SHGC:

WindowMaterial:SimpleGlazingSystem,
  Nonres Window Glazing Layer,           !- Name
  1.99,              !- U-Factor {W/m2-K}
  0.42,                     !- Solar Heat Gain Coefficient
  0.4;                     !- Visible Transmittance

The energy consumption is supposed to decrease, but not. The total energy including total site energy, net site energy, total source energy, and net source energy have increased. I checked the end-use table, and I found that after the modification, the load for heating, cooling, fans, and pump has increased. But they are supposed to be decreased.

I tried to keep the SHGC and only modified the U factor:

WindowMaterial:SimpleGlazingSystem,
  Nonres Window Glazing Layer,           !- Name
  1.99,              !- U-Factor {W/m2-K}
  0.4,                     !- Solar Heat Gain Coefficient
  0.4;                     !- Visible Transmittance

I ran the simulation again, and the energy consumption has reduced.

The only difference between the two modified models is SHGC. For Rochester, NY, which is the cold climate zone, the energy consumption is supposed to be decreased if increasing the SHGC I was wondering if it's may cause by the constraints of EP+? Or some settings in the model will lead to this issue?