Establish Program WCET and Energy Consumption Prediction Model Based on L-M Algorithm

In some hard real-time systems, the system has high requirements for program execution time and energy consumption. If the program runs overtime or the energy is exhausted in advance, it will have a significant impact on system security. In order to be able to more accurately predict the WCET and energy consumption of the program and provide support for the subsequent search for the best optimization method that optimizes WCET and the average execution time at the same time, this paper gives a set of feasible methods that can predict the worst execution time and average execution time of the program. On the basis of existing research, the static method of program execution time estimation is integrated with the dynamic method, and the WCET and energy consumption of the program are estimated using the sample program features such as dynamic instruction features, and the L-M (Levenberg-Marquardt) algorithm is used to train neural network. And compared with the traditional regression algorithm, add quantitative indicators and verify the feasibility of the method. The method in this paper can make an accurate prediction of the execution time of the program. The research is helpful to the follow-up development of this field and provides a useful reference and reference for the further optimization of the program.