Neural network based correlations for estimating temperature elevation for seawater in MSF desalination process

Modelling played an important role in simulation, optimisation, and control of multi-stage flash (MSF) desalination processes. Top brine temperature (TBT) is one of the many important parameters that affect optimal design and operation of MSF processes. Within the MSF process model, calculation of TBT is therefore important. For a given pressure, TBT is a function of boiling point temperature (BPT) at zero salinity and temperature elevation (TE) due to salinity. In this work, we develop several neural network (NN) based correlations for predicting TE. It is found that the NN based correlations can predict the experimental TE very closely. Also predictions by the NN based correlations were good when TE values, obtained using existing correlations from the literature are compared. Due to advancement of the microcomputer, plant automation becomes reliable means of plant maintenance. NN based correlations (models) can be updated in terms of new sets of weights and biases for the same architecture or for a new architecture reliably with new plant data.

In this work, several NN based correlations for predicting TE of seawater is developed. For each correlation, a multi-layered feed forward Table 14 Comparison of prediction by NN_Cor_2 and experimental data from Fabuss data Range: x = <1.5–13> BPT = <37–121.11> Salinity BPT TEEXP TENN Salinity BPT TEEXP TENN Salinity BPT TEEXP TENN 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 0.144 0.157 0.170 0.183 0.196 0.209 0.222 0.235 0.248 0.195 0.212 0.229 0.247 0.264 0.282 0.299 0.317 0.334 0.247 0.269 0.291 0.313 0.335 0.357 0.379 0.401 0.423 0.355 0.387 0.418 0.449 0.481 0.512 0.544 0.575 0.607 0.099 0.116 0.133 0.150 0.164 0.178 0.191 0.206 0.224 0.141 0.162 0.183 0.204 0.224 0.243 0.261 0.280 0.302 0.186 0.209 0.234 0.259 0.284 0.309 0.333 0.357 0.384 0.287 0.315 0.345 0.378 0.412 0.447 0.483 0.518 0.555 4.5 4.5 4.5 4.5 4.5 4.5 4.5 4.5 4.5 5.5 5.5 5.5 5.5 5.5 5.5 5.5 5.5 5.5 0.412 0.448 0.484 0.520 0.556 0.593 0.629 0.666 0.702 0.470 0.511 0.552 0.593 0.634 0.675 0.716 0.758 0.790 0.529 0.575 0.621 0.667 0.713 0.759 0.805 0.852 0.898 0.589 0.640 0.691 0.743 0.794 0.845 0.896 0.948 0.999 0.344 0.374 0.407 0.443 0.481 0.520 0.561 0.603 0.645 0.405 0.437 0.473 0.511 0.553 0.597 0.643 0.690 0.738 0.471 0.505 0.543 0.585 0.630 0.678 0.728 0.780 0.834 0.541 0.578 0.618 0.663 0.711 0.763 0.818 0.875 0.934 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 6.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 7.5 0.651 0.708 0.764 0.820 0.876 0.933 0.989 1.046 1.103 0.715 0.776 0.838 0.899 0.961 1.022 1.084 1.146 1.208 0.780 0.846 0.913 0.980 1.047 1.114 1.181 1.248 1.315 0.846 0.918 0.990 1.062 1.135 1.207 1.279 1.352 1.425 0.613 0.654 0.698 0.745 0.797 0.853 0.912 0.974 1.038 0.687 0.732 0.780 0.832 0.888 0.947 1.011 1.077 1.146 0.761 0.811 0.865 0.921 0.982 1.046 1.114 1.185 1.259 0.834 0.891 0.950 1.013 1.079 1.148 1.221 1.297 1.377 Maximum deviation 0.1531 Table 15 Comparison of prediction by NN_Cor_3 and experimental data from Bromley data Range: x = <1.5–7.136> BPT = <60–120> Salinity BPT TEEXP TENN Salinity BPT TEEXP TENN 1.57 1.69 2.25 3.05 3.74 4.53 5.18 5.76 6.42 7.14 1.68 1.71 2.15 3.03 3.35 3.47 3.62 4.28 4.88 5.06 5.06 5.45 6.03 6.52 6.59 6.62 6.72 6.74 7.10 7.14 0.178 0.192 0.258 0.356 0.440 0.542 0.629 0.709 0.803 0.907 0.248 0.251 0.318 0.454 0.505 0.524 0.547 0.657 0.756 0.789 0.790 0.857 0.960 1.049 1.061 1.062 1.088 1.090 1.157 1.167 0.172 0.187 0.261 0.367 0.457 0.560 0.644 0.723 0.820 0.943 0.253 0.257 0.325 0.468 0.521 0.541 0.566 0.678 0.781 0.813 0.814 0.883 0.992 1.089 1.104 1.111 1.131 1.135 1.216 1.225 1.85 2.54 2.64 3.47 3.47 4.03 4.78 5.39 5.84 5.91 6.32 6.41 6.57 6.58 6.75 6.99 1.95 2.54 2.60 3.06 3.35 3.91 4.42 4.91 5.40 5.93 6.41 6.59 6.87 0.241 0.331 0.347 0.461 0.461 0.542 0.654 0.749 0.816 0.830 0.894 0.910 0.937 0.936 0.964 1.008 0.324 0.425 0.437 0.516 0.567 0.670 0.764 0.856 0.951 1.056 1.154 1.193 1.246 0.243 0.343 0.359 0.481 0.482 0.565 0.679 0.773 0.844 0.856 0.927 0.942 0.972 0.973 1.005 1.054 0.329 0.430 0.441 0.523 0.577 0.681 0.779 0.875 0.974 1.083 1.190 1.233 1.298 Maximum deviation network trained with back propagation method is used. The NN based correlations can predict the experimental TE and TE obtained by using correlations form literature very closely. The proposed NN model structure (with one hidden layer and four neurons in hidden layer) is capable of predicting the experimental TE very closely. For a given architecture, any correlation can be updated with additional data from other sources or a new correlation can be developed for 0.0588 the new source data. Predictions by different NN based correlations (for different salinity and BPT) within the training range follow the expected trends and it is within the engineering accuracy. It is found that redevelopment of an NN based correlation form an existing one with new data is very easy. However, the prediction of TE by the new correlation solely depends on the source of data (i.e., the seawater constituents and conditions and above all, accuracy of the data).