It is found by experiment that under the thermal convection condition, the temperature fluctuation in the urban canopy layer turbulence has the hard state character, and the temperature difference between two points has the exponential probability density function distribution. At the same time, the turbulent energy dissipation rate fits the log-normal distribution, and is in accord with the hypothesis proposed by Kolmogorov in 1962 and lots of reported experimental results. In this paper, the scaling law of hard state temperature n order structure function is educed by the self-similar multiplicative cascade models. The theory formula is Sn = n/3μ{n(n+6)/72+[2lnn!-nln2]/2ln6}, and μ Is intermittent exponent. The formula can fit the experimental results up to order 8 exponents, is superior to the predictions by the Kolmogorov theory, the β And log-normal model.