The neural networks
 |
To model the collector array as simply as possible, a very small network is used. Using a unique non linear neuron on the hidden layer and a linear output neuron leads to good results when considering modeling the array when the fluid is flowing during two consecutive hours; which means that the system is in a quasi steady state and that it is not necessary to use complex models as presented in [9]. This is also similar to the conclusion given in [10] stating that parameter estimation of a solar collector array is reliable when high temperatures are obtained within the collectors. The inputs are the inlet temperature T4, the global radiation, the ambient temperature. The output is the outlet temperature T1. Figure 3 shows the differential (in %) between estimated values and actual values when the system is stable all year long. Due to the fact that the system is more likely to run two consecutive hours in summer, the error is lower during these months. Nevertheless, the maximum error is less than 1%. The equation representing the estimated values
For the connecting pipes a similar network is used, the inputs are the inlet temperature of the pipe, the ambient temperature, and the global radiation. The output is the outlet temperature of the pipe. In this case the regression R value are closer to unity, the difference is -2.6 10-6, and the equation is
Test = 0.9997 Tact - 0.0174
