This paper proposes fractional order graph neural networks (FGNNs), optimized by the approximation strategy to address the challenges of local optimum of classic and fractional graph neural networks which are specialised at aggregating information from the feature and adjacent matrices of connected nodes and their neighbours to solve learning tasks on non-Euclidean data such as graphs. Meanwhile the approximate calculation of fractional order gradients also overcomes the high computational complexity of fractional order derivations. We further prove that such an approximation is feasible and the FGNN is unbiased towards global optimization solution. Extensive experiments on citation networks show that FGNN achieves great advantage over baseline models when selected appropriate fractional order.