Aiming at the problem of humidity sensor data distortion caused by temperature changes in utility tunnel, an improved Levy flight particle swarm optimization algorithm (ILPSO) is proposed to compensate data errors. Firstly, a neural network with prediction error is established, and the initial parameters of the network are found by PSO. Then, an improved levy flight is added in the PSO search process, the probability of particle flight is inversely proportional to the distance to the optimal particle, and when approaching the optimal particle, it reversely escapes from the optimal particle with a larger probability, thus overcoming the problem of particle premature. Finally, the network retrains with the output of PSO as the initial parameter. In the sensor error test experiment and stability experiment, the compensation effect of ILPSO algorithm is the best, the humidity value error after compensation is less than 5%, MSE is the lowest, and the stability is the best. In summary, compared with the traditional Levy flight, ILPSO algorithm has stronger adaptability to error prediction network, faster convergence, and improves the accuracy and stability of humidity sensor temperature compensation.