There is a dataframe df_heat with heat_output over temp. As a first approximation you may fit the curve with two linear model and one breaking point:
lm(data = df_heat, heat_output ~ temp + I(pmax(17-temp, 0)))
I like to fit a linear model for low temperatures and and a sigmoid function for higher temperatures like suggested in this paper (german). The breaking point can be a fixed value at 5 °C.
How can I specify this formula in R?
A model is proposed below. The equation involves four parameters a,c,tc and lambda. The shape of the curve is controled by the parameter lambda. Non-linear regression can be used to evaluate the four parameters and fit to data.