Abstract: Due to the differences in the ability of different operating parameters of equipment to reflect its degree of abnormality, the identification results of equipment abnormalities affect the quarterly growth rate of the minimum annual cost of equipment after maintenance. Therefore, a study on abnormal maintenance of thermal power plant equipment operation based on GA improved BP neural network algorithm was carried out. Optimize the input feature values, weights, and thresholds of BP neural network using GA. The optimized BP neural network outputs feedback on the degree of equipment operation abnormalities related to different operating parameters and the degree of equipment abnormalities, thereby ensuring the reliability of anomaly recognition results. Based on the degree of equipment abnormality, the annual minimum cost of the current and repaired equipment was calculated separately, and maintenance measures were decided based on the relationship between the two. In the test results, the design method effectively identified the abnormal state of the equipment and maintained a stable annual increase in the degree of equipment abnormality after maintenance, with the quarterly increase in the minimum annual cost remaining within 8000 yuan.