Modern industrial power plants require continuous monitoring to ensure their safe and reliable operation. This monitoring depends heavily on accurate sensor readings from a large number of sensors, making sensors an important part of any plant. The possibility of more than a single sensor being simultaneously faulty should not be overlooked. In this paper, a nonlinear system data validation approach based on a robust input-training neural network is proposed for detecting, isolating, and reconstructing multiple sensor faults in a power plant. An influence factor function and reliability coefficients were introduced into an objective function for the purpose of inhibiting the influence of numerous failure data with significant errors. The proposed method was evaluated on a single-shaft gas turbine from a natural gas combined cycle power plant. The result shows that the proposed model has a high degree of accuracy in multiple sensor failure cases.