Abstract: BACKGROUND: Optimal power load strength training can effectively increase the output power of skeletal muscle, promote health and improve sports performance. However, how to quickly determine the optimal power load is often a difficult problem in the practice of strength training, and is also a hot topic in the research of scholars at home and abroad.
OBJECTIVE: To study the mathematical relationship between maximum strength, height, weight and optimal power load by using back propagation neural network modeling, so as to build a model to predict the optimal power load. 
METHODS: Fifty-two subjects (46 subjects for test, 6 subjects for forecast) were recruited. The maximum strength test and maximum power output test were carried out on the subjects to construct the optimal power load forecasting model based on error back propagation correction training algorithm, and the trained back propagation neural network model was used to predict the optimal power load in the new sample to explore the prediction effect of the model. 
RESULTS AND CONCLUSION: Using the strong self-learning and reasoning ability of back propagation neural network, the optimal power load forecasting model was constructed with 3 input layers, 10 hidden layers and 1 output layer. In terms of the prediction accuracy of different strength training methods, the mean relative error of bench press throw and half squat is 9%, and the mean absolute error is 3.79 kg and 6.91 kg respectively. Back propagation neural network prediction method can effectively predict the optimal power load, which makes the determination method of optimal power load more diversified and intelligent.

Key words: power, output power, optimal power load, neural network, prediction method, squat, bench press