The problem of load identification denotes identifying loads based on the measurement of structural responses， which is the inverse problem in structural dynamics. A load identification method based on time-delay neural network is proposed in this pa‐ per， and numerical examples based on simulation and experiments are provided to show that the method overperforms normal backpropagation neural network in accuracy of identification. Additionally， statistic pooling is introduced on the basis of the method， and it is proved that the method performs well in noisy environment compared with BP neural networks. based on the load identifi‐ cation methods mentioned above， a sensor placement optimization based on particle swarm optimization algorithm is proposed， and the optimal sensor placement is able to reduce the error of identification by 90% compared with the random sensor placements， meanwhile the minimum spacing of installation among sensors is also ensured during the optimization.