<p>An effective energy-efficiency hybrid precoding approach is developed based on cross-entropy optimization (CEO) for dynamic fully-connected (DFC) structure in this paper, named as PS-CEO algorithm. The DFC structure is introduced to improve energy efficiency (EE), which adaptively controls the connection state between antennas and radio frequency (RF) chains by switches according to the channel state information. Firstly, by using the redefined array gain efficiency, the EE maximization problem with multiple constraints is transformed into the array gain efficiency maximization problem with switching constraints. And then, using the theory of CEO, the switch precoding matrix optimization of array gain efficiency maximization problem is converted as a optimization problem of {0,1} probability distribution. On the basis, the switch precoding matrix feasible sample can be generated by {0,1} probability distribution parameters. The “high quality” samples are selected by the array gain efficiency maximization criterion. Then, through minimizing the cross entropy between the estimated and expected values of the probability distribution parameters, the probability distribution parameters are updated. Therefore, the optimal switching precoding matrix is acquired by the aforementioned CEO theory. Finally, a block diagonal technique is used to obtain the digital precoding matrix. Theoretical and simulations analyses show that the EE of the proposed PS-CEO algorithm is superior to other algorithms while obtaining satisfactory SE with lower computational complexity. Specially, the EE of the proposed method is at least about 0.33 bits/Hz/J and 0.29 bits/Hz/J higher than that of the superior WP-AM method with signal-to-noise and the number <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(N_{RF}\)</EquationSource> </InlineEquation> of RF chains.</p>