Non-orthogonal multiple access (NOMA) and cognitive radio technology are considered to be the two most promising techniques of the new generation mobile communication system for improving spectral efficiency. In this paper, to optimize the efficiency of the power domain NOMA cognitive radio network, a sub-user power allocation model is established to meet the minimum system throughput and the maximum interference of the main user. Then, the sub-channel throughput formula is decomposed to obtain the sub-channel power allocation coefficient and sub-channel power consumption rate. For the first problem, the objective function is equivalent to two convex function difference forms by using the difference of convex (DC) programming algorithm, and the non-convex problem is transformed into the convex optimization problem by using the first-order Taylor expansion continuous approximation, to obtain the optimal power distribution coefficient of the sub-user of the subchannel multiplexing. To solve the second problem, Dinkelbach, subgradient algorithm and Lagrange function are used to obtain the optimal subchannel power consumption rate. The simulation results show that the proposed power allocation algorithm has fast convergence speed and low time complexity, and its average system energy efficiency performance is far better than the fractional power allocation algorithm.