Hydrogen solubility in heavy n-alkanes; modeling and prediction by artificial neural network

In this manuscript, solubility of hydrogen in heavy n-alkanes (C10H22, C16H34, C28H58, C36H74 and C46H94), within temperature range of 283–448 K and pressure range of 1.15–15.97 MPa, has been investigated. Due to unreliable estimations of classical techniques and complexity of proposed accurate methods, artificial neural network (ANN) technique has been used to model and predict hydrogen solubility in heavy n-alkanes. In first stage, separate networks have been designed for studied systems. Temperature and pressure are inputs of each network and hydrogen solubility is the output. Though precise modeling results, this traditional ANN method has several disadvantages. Beside large amount of adjustable parameters, application of each designed network is restricted to the system which it has been trained for. In second stage, one 3-4-1 network has been designed for all studied systems and number of carbon atoms in n-alkane has been used as a new input. This new strategy results in a general simple model based on molecular structure. Average relative deviation of 3-4-1 network has been calculated 1.66%. In addition to accurate modeling, designed network can predict hydrogen solubility in heavy n-alkanes based on available data of similar systems with acceptable accuracy. This capability is very important especially when experimental data is not available. Other methods, including traditional equations of states, SAFT theory, modified forms of Henry's law and traditional modeling via ANN, do not have such a capability.

► Accurate modeling of hydrogen solubility in heavy n-alkanes via ANN.
► Accurate hydrogen solubility prediction based on available data of similar systems.
► Indicating superiority of designed ANN over traditional ANN method.
► Indicating superiority of designed ANN over SAFT models.
► Indicating superiority of designed ANN over traditional techniques.