For the hidden layer structure and learning speed of the process neural network, a structure-adapted extreme process neural network is proposed based on the extreme learning machine.Firstly, the adaptive model structure is achieved by incrementally adding the process neuron until output error meets accuracy requirements.Then, in order to eliminate the redundant nodes, a correlation discrimination method is proposed, in which Gram-Schmidt orthogonalization is executed to new node output.Finally, a quantum-inspired cuckoo search algorithm is proposed, which is used to optimize the parameters generated by input weight function orthogonal basis expansion.The effectiveness is validated by the comparative analysis based on Mackey-Glass and shale oil TOC prediction.The simulation result shows that the approximation efficiency and training speed of the proposed model are obvious improved.