Design of Neural Network Gain Scheduling Flight Control Law Using a Modified PSO Algorithm Based on Immune Clone Principle

Response to the question that the traditional gain scheduling which is one-parameter adjusting is complex and difficult to find suitable adjusting rule and so the flying qualities in full flight envelope curve specially for those flight conditions between the operating points can´t be guaranteed for a modern fly-by-wire flight control system, a design method of three-layer BP network gaining-scheduling for multi-parameters in full flight envelope curve is proposed. Through a comprehensive analysis of PSO algorithm, immunity clone algorithm was introduced to the PSO algorithm based on traditional velocity-displacement operator. A modified global PSO algorithm based on immune clone principle (ICMPSO) is also developed to optimize network weights to improve approximation for nonlinear functions to overcome disadvantages that BP neural network easily involves in a slow convergence and local extremum using gradient descent method to train network weights. The outputs are optimal feedback gains in operating points with ICMPSO method based on FCS optimizing strategy of reference model. Double-parameters gain-scheduling mechanism according to Mach number and dynamical pressure is effectively realized with proposed method. The results show that the advanced algorithm can greatly improve training speed and precision of BP neural network. It not only guarantees good system response at the designed flight conditions but gives more finely carve up result of control parameters between operating points with the neural network gain-scheduling controller. Meanwhile, it also gets a good control effect for the system with about 20% modeling error. The method is of some enlightening and referenced value to engineering application.