Research On Chaotic Neural Network Optimization Algorithm

Neural network provides a powerful tool for solving problems such as classification recognition,fault diagnosis and path planning.The neural network uses the gradient descent method,and the direction of each descent is the direction of the fastest decline of the function.However,the local gradient on the error curve usually does not point to the global minimum,and the convergence path zigzag.So gradient descent is easy to converge to a local minimum.Applying the ergodic property of chaos theory to neural networks can effectively avoid the local minimum problem of traditional neural networks,which brings a new development direction to the research of neural networks.Based on chaos theory and neural network,this paper proposes a new chaotic neural network and its optimization algorithm.These new optimization algorithms have been applied to classification recognition,fault diagnosis and path planning.The main innovations and work are as follows:(1)A chaotic neural network based on complex Logistic mapping was proposed and applied to ECG classification and recognition.Firstly,the bifurcation diagram of complex Logistic chaotic map was drawn,and the Lyapunov index was calculated to analyze its chaotic characteristics.Then,the weight was calculated by complex Logistic chaotic map,and the global optimization ability of complex chaotic neural network was enhanced.Finally,MIT-BIH database was used for ECG classification to verify the effectiveness of the proposed algorithm.The simulation results showed that the complex chaotic neural network can improve the accuracy of ECG recognition.(2)A convolutional neural network based on chaotic activation function is proposed and applied to bearing fault diagnosis.Firstly,a new activation function of Logistic chaotic mapping is proposed,which has the ergodic property and pseudo-randomness of chaos.Then a new Logistic chaotic mapping activation function is introduced in the fully connected layer of the convolutional neural network.Through the pseudo-randomness and ergodicity of chaos,the neural network can jump out of the local minimum point with a certain probability in the early stage and find the approximate global minimum point in the late stage of gradient descent.Finally,the data sets of Western Reserve University and American Society for Mechanical Failure Prevention Technology were used to verify the effectiveness of the algorithm.Simulation results showed that the algorithm has high recognition accuracy and good robustness.(3)A hybrid optimization algorithm based on transient chaotic neural network and genetic algorithm was proposed and applied to route planning of travel dealers and wind farm operation and maintenance.Firstly,chaotic neural network was used for the first step of path planning.The programming results were encoded as chromosomes to replace the least fit individuals in the genetic algorithm population.Then,according to the principle of survival of the fittest,the population was selected,hybridized and mutated to guide the circular evolution and obtain a new path.Finally,the travel salesman problem and remote wind farm path planning problem were used to verify the effectiveness of the proposed algorithm.The results showed that the proposed algorithm generates a shorter optimal path than other algorithms.