Overview: In recent years, the number of new photoelectric measurement equipment has increased rapidly, the composition has become more and more complex, the accuracy has gradually improved, and the functions have become more comprehensive. During the normal life cycle of large-scale optoelectronic measurement equipment, engineers seek to maintain the performance of the equipment with the lowest possible cost and as few personnels as possible, so the demand for research on failure prediction and diagnosis technology is increasing. The traditional on-site manual diagnosis and maintenance method requires a lot of manpower and material resources, and it takes a long time to complete a test and diagnosis. The accuracy of the diagnosis is very dependent on the familiarity and experience of the operator. Once a fault occurs, it is difficult to quantify the time for positioning and troubleshooting, which affects the combat effectiveness of the equipment. In fact, major faults that affect the performance of equipment are generally easy to repair in the early stage, but often due to incomplete detection and diagnosis methods, they cannot be detected or cannot be detected on-site in time, resulting in major faults accumulated over time. In the fault diagnosis of photoelectric measurement system, the prediction of tracking error is particularly important. CS-BP algorithm has strong self-adaptive and self-learning ability, and can obtain more reliable results without additional human intervention, so it is often used for fault diagnosis and parameter prediction of large-scale systems. Based on the BP neural network, this article uses the cuckoo algorithm to optimize the threshold and weight, and proposes a CS-BP algorithm. This essay uses the azimuth guidance, pitch guidance, azimuth encoder, pitch encoder and time data of the photoelectric measurement system to predict the tracking error. Compared with the traditional neural network algorithm, the algorithm utilizes the cuckoo's excellent feature of finding extreme values, and solves the problem that the neural network algorithm cannot obtain the optimal solution due to improper initial threshold and weight settings. The experimental results show that compared with the traditional BP neural network and the BP neural network optimized by the genetic algorithm (GA-BP), the number of iterations of the CS-BP algorithm is 21 and 60 times less, and the average relative error of the prediction is 4.85% and 1.57% lower, respectively. Therefore, CS-BP algorithm has a faster convergence speed and higher prediction accuracy, and is suitable for application in fault diagnosis of optoelectronic measurement systems.In recent years, with the increasing number and complexity of photoelectric measurement systems, the demand for fault diagnosis is also increasing. In the fault diagnosis of the photoelectric measurement system, the prediction of its tracking error is particularly important. In this paper, we propose a BP neural network algorithm optimized by the Cuckoo algorithm (CS-BP). The tracking error can be predicted by using the azimuth guidance, pitch guidance, azimuth encoder, pitch encoder and time data of the optoelectronic measurement system. Compared with the traditional neural network algorithm, this algorithm uses the excellent characteristics of Cuckoo to find the extreme value, and solves the problem that the neural network algorithm cannot get the optimal solution due to the improper setting of the initial threshold and weight. The experimental results show that, the number of iterations with CS-BP is 21 and 60 less than the traditional BP neural network and the BP neural network optimized by the genetic algorithm (GA-BP), respectively. The relative errors are 4.85% and 1.57% lower, respectively. Therefore, the CS-BP algorithm has a faster convergence speed and higher prediction accuracy, and it is suitable for fault diagnosis of photoelectric measurement system.