Figure 4
The common inputs are pile length, pile diameter, initial temperature, average soil thermal conductivity, concrete thermal conductivity, fluid velocity and number of loops. The Power Output Surrogate also uses inlet temperature. Its sequence contains a dense layer with 1,024 features and rectified linear unit activation, batch normalisation and dropout of 0.2. This is followed by dense layers with 256 and 128 features, each with rectified linear unit activation, batch normalisation and dropout of 0.2. The final linear dense layer has 17 features. The Thermal Response Surrogate also uses heat flux. Its first two dense layers each have 512 features and rectified linear unit activation, followed by batch normalisation and dropout of 0.3. A third dense layer has 256 features with the same activation, normalisation and dropout. The final linear dense layer has 2 times 120 features.ANN surrogate models architecture