Support Vector Regression-Based Performance Prediction for Unmanned Aerial Vehicles

The increasing incidence of fires has resulted in a growing number of casualties and economic losses. On the one hand, traditional firefighting approaches and equipment are inefficient and expose firefighters to great danger. On the other hand, in recent years, unmanned aerial vehicles (UAVs) are gradually being applied in firefighting fields such as fire detection and extinguishing due to their safety, flexibility, and low cost. The performance of the firefighting UAV has a significant impact on its efficiency in performing their tasks. To calculate the performance quantitatively, support vector regression (SVR) is employed to establish the performance prediction model. However, it is difficult to set SVR parameters efficiently because of the high dimensionality, nonlinearity, and complexity in this problem, which restricts the prediction effect. Therefore, the mutant shuffled frog leaping algorithm (MSFLA) based SVR method is proposed to optimize SVR parameters and predict the performance of the UAV. First, the step size equation is modified in MSFLA, which expands the search range in the iterations and can help to find better solutions. Meanwhile, considering that traditional optimization algorithms are prone to falling into local optimums, the Cauchy mutation is introduced into the position update process of the proposed MSFLA to overcome this shortcoming. The mutation strategy improves the global optimization capabilities. Finally, the performance prediction based on the proposed methods are simulated. The simulations focus on comparing the prediction errors of the different methods on the testing set. The average predicted errors and four commonly used assessment indicators are calculated, and the corresponding results suggest that the prediction effectiveness of MSFLA-SVR is much better than the other methods. Moreover, the results of statistical test further demonstrated the superiority of the proposed approaches.