To compare the performance between methods, the relative error was computed.
Table 1 lists the maximum absolute errors E, which resulted from the application of GJGOMM, for various values of N, while in Table 2 we display a comparison between the relative errors obtained by the application of the two methods namely, the first-order method (1OM) and the second-order methods (2OMs) developed in [35] with the relative errors resulting from the application of GJGOMM.
A shorter 5-day time interval (on average) between samplings lowers the relative error in the SPM fluxes to below 10%, a result directly related to the increased number of Q C pairs available for fitting.
As we can see from Table 4, this novel identification model based on the basic genetic algorithm has a relatively higher precision: the relative error between identification results of pollutant amount and analytic results varied from −1.7 to 5.0 %, and the relative error between identification results of source position and analytic results varied from 0 to −1.6 %.
The RMSE values for both TRNSYS & eQUEST were found varying between 2.1 12.5% and the Relative Error was found varying −19.6%to18.0%0% respectively.
However, as the receive antenna trajectory is kept constant between different interpolation factors, the relative error is mainly influenced by the repeatability of the setup, the noise, and the interpolation itself.
The MOBP network converged after 3900 times of training as shown in Fig. 7, with initial weights set randomly between −1 and 1, the relative error for convergence threshold set to.01, and the momentum factor set equal to.8.
The relative error between the theoretical and the generated signals is 1.3 × 10-2, and the relative error between the estimated and the generated signals is 2.2 × 10-5.
The relative error between these real migrations and our predictions was between 4% and 22%.
