A frequency item mining based embedded feature selection algorithm and its application in energy consumption prediction of electric bus

In the engineering practice of applying embedded feature selection algorithm to construct EV energy consumption prediction model, the constructed regression learners are often affected by random factors appeared in the process of data set sampling, algorithm initialization, computing platform resource scheduling and so on, which makes the prediction results of multiple regression learners constructed with the same feature combination different. This seriously affects the optimization process of energy consumption prediction model, resulting in the failure to find the optimal feature combination, and reduces the accuracy of the prediction results. To solve this problem, an embedded energy consumption prediction model construction method based on frequency item mining and evolutionary computing was proposed. In this algorithm, the combination of input characteristic variables is regarded as individual in the population, the prediction result of regression model is regarded as the fitness function, and the randomness of fitness function is corrected online by the statistical results of frequency items. Simulation results show that the algorithm solves the interference of randomness appeared in the process of resource scheduling, class library function reference, data set segmentation, etc., ensures the stability of feature combination in the optimization process of the prediction model, and gets accurate prediction results. •Random factors lead to the uncertainty of energy consumption prediction model.•Random factors restrict the application of embedded feature selection algorithm.•Frequency item can modify embedded feature selection algorithm's fitness function.•Frequency item mining can improve the accuracy of energy consumption prediction model.