SummaryIndustrial wireless sensor networks (IWSNs) are very important to improve and simplify the way to manage, monitor, and control industrial factories. IWSNs have many benefits, but because of their vulnerability to extremely complex and variable industrial contexts, they also limit some of the potential that is currently available and present difficulties on numerous fronts. The proposed work develops energy‐efficient routing technique to minimize the overall energy consumption. Heterogeneous mobile nodes are randomly deployed in the experimental region; then, region‐based grid formation is done in the proposed method. In each grid, the base node is selected utilizing the parameters like residual energy and distance of mobile nodes. Hybrid COOT‐HOA optimization is used in the proposed method for selecting the optimal base node in each grid. A deep learning‐based machine learning algorithm called long short‐term memory (LSTM) is utilized to predict the future direction of movement (DOM) of each mobile node in the grid. Estimation of signal parameters via rotational invariance technique (ESPRIT) is utilized to find an active zone from the sender node to direction of base node. Then, the sender node transmits the data to its nearest node in the active region. This proposed energy‐efficient routing algorithm is tested with several metrics which attains better performance like 94% packet delivery ratio, 7% packet loss, average residual energy of 9.5 J, and 3.4 Mbps throughput. Thus, the energy‐efficient routing protocol used in the proposed approach transfers the data in an energy‐efficient manner for IWSN.