In the digital age, processing and transmitting the IoT data collected in network is one of the most essential technologies of the wireless sensor networks. Among all data processing models, the clustering is the essential tasks. In this paper, the novel stochastic neural network optimization based IoT data clustering algorithm is studied. The proposed model contains various novelty: First, as the PNN model layer adopts the radial basis kernel function (KF), which fully considers the interweaving effect between samples of different categories and has excellent fault tolerance, the novel Cuckoo Search (CS) is applied to make the neural network stronger. In addition, the algorithm is extended by so-called Lévy flights instead of a simple isotropic random walk. Second, CS can explore the search space more efficiently than algorithms using standard Gaussian processes. This advantage combines local and search capabilities with the guaranteed global convergence. Third, mathematically, the IoT data is expressed as the potential function, from an energy perspective, the data field uniformly expresses various methods of simulating physical forces in terms of the potential values. The experimental results show that the proposed algorithm can obtain the better clustering accuracy compared with the control sets. Our method proves better robustness to the iterations.