This study proposes a mining equipment anomaly detection model based on bidirectional long short memory network and improved subtraction average optimizer, aiming to solve the problems of low efficiency and poor accuracy of traditional equipment anomaly detection methods. This model dynamically weights key features by introducing attention mechanism, combines convolutional neural network to extract local temporal features, and uses guided regularized random forest for feature selection to reduce data dimensionality and noise interference. Furthermore, the improved subtractive averaging optimizer utilizes chaotic mapping initialization and golden sine step size adjustment to enhance the convergence speed and stability of the model.t. The results show that on the multi-sensor time series dataset of mining equipment and the fault simulation dataset of mining machinery, the accuracy rate of the proposed model reaches 94.17% and 95.48% respectively, and the F1 scores are 93.02% and 95.57% respectively, which is superior to the existing mainstream detection models. In the actual application test, the detection accuracy of the model has been improved to 97.32%, with an average detection time of only 0.39 seconds, verifying its efficiency and practicality. Research has shown that the model can provide high-precision and high-efficiency solutions for abnormal detection of mining equipment, which is of great significance for ensuring the safety of mining production.