In order to solve the problem that the leakage of privacy content of images in the process of public channel transmission results in endangering information security, a new encryption method of greyscale image was proposed. The iteration of coupled logistic map was used to generate two-dimensional chaotic sequences. One of the sequences was used to generate the coefficients of cat map. The another was used to scramble the pixel positions. The traditional image encryption method based on cat map was used to encrypt the image pixel position, while the proposed encryption method was used to adopt different cat map coefficients for different pixel groups, so as to transform the grey value of each pixel in the group. In addition, bidirectional diffusion was adopted by the method to improve the security performance. The proposed method has simple encryption and decryption processes, high execution efficiency, and no limitation for the image size. Security analysis shows that the proposed encryption method is very sensitive to secret keys, and has good stability under multiple attack methods.

Aiming at the imbalance of training data set, an integrated support vector machine classification algorithm was proposed by combining sampling method with ensemble method. Firstly, unsupervised clustering was performed on an unbalanced training set, then the underlying local attention support vector machine was used to partition the data set so as to precisely control the local features of data sets. Finally, top support vector machine was used to predicte classification. The evaluation results on UCI dataset show that compared with the popular algorithms such as sampling based Kernelized Synthetic Minority Over-sampling TEchnique (K-SMOTE), integration based Gradient Tree Boosting (GTB) and cost sensitive ensemble algorithm (AdaCost), the proposed support vector machine algorithm can significantly improve the classification effect and solve the problem of unbalanced data set to a certain extent.

Aiming at the shortcomings of large feature set dimensionality, data redundancy and low fault recognition rate in existing fault diagnosis method based on simple combination of multi-classes features, a fault diagnosis method based on heterogeneous feature selection and fusion was proposed. The clustering characteristics of the feature data was analyzed according to the contours of the data of various class of features, and the redundant feature dimensions which are weakly clustered and not useful for fault classification were removed, only the feature dimensions with strong clustering characteristics were retained for the fault recognition. In the bearing fault diagnosis experiment, time-domain statistics and wavelet packet energy of fault signals were optimally selected and merged, and Back Propagation (BP) neural network was used for fault pattern recognition. The fault recognition rate reached 100%, which is significantly higher than that of the fault diagnosis method without feature selection and fusion. Experimental results show that the proposed method is easy to be implemented and can significantly improve the fault recognition rate.

Since fractional order chaotic time series prediction has low precision and slow speed, a prediction model of new orthogonal basis neural network based on Quantum Particle Swarm Optimization (QPSO) algorithm was proposed. Firstly, on the basis of Laguerre orthogonal basis function, a new orthogonal basis function was put forward combined with the neural network topology to form a new orthogonal basis neural network. Secondly, QPSO algorithm was used for parameter optimization of the new orthogonal basis neural network, thus the parameter optimization problem was transformed into a function optimization problem on multidimensional space. Finally, the prediction model was established based on the optimized parameters. Fractional order Birkhoff-shaw and Jerk chaotic systems were taken as models respectively, then chaotic time series produced according to Adams-Bashforth-Moulton estimation-correction algorithm were used as the simulation objects. In the comparison experiments on single-step prediction with Back Propagation (BP) neural network, Radical Basis Function (RBF) neural network and general new orthogonal basis neural network, Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) of the new orthogonal basis neural network based on QPSO algorithm were significantly reduced, and Coefficients of Decision (CD) of it was closer to 1; meanwhile, Mean Modeling Time (MMT) of it was greatly shortened. The theoretical analysis and simulation results show that the new orthogonal basis neural network based on QPSO algorithm can improve the precision and speed of fractional order chaotic time series prediction, so the prediction model can be easily expanded and applied.

Concerning low precision and slow speed of traditional intelligent optimization algorithm for parameter identification in chaotic system, a new method of parameter identification in chaotic system based on feedback teaching-learning-based optimization algorithm was proposed. This method was based on the teaching-learning-based optimization algorithm, where the feedback stage was introduced at the end of the teaching and learning stage. At the same time the parameter identification problem was converted into a function optimization problem in parameter space. Three-dimensional quadratic autonomous generalized Lorenz system, Jerk system and Sprott-J system were taken as models respectively, intercomparison experiments among particle swarm optimization algorithm, quantum particle swarm optimization algorithm, teaching-learning-based optimization algorithm and feedback teaching-learning-based optimization algorithm were conducted. The identification error of the feedback teaching-learning-based optimization algorithm was zero, meanwhile, the search times was decreased significantly. The simulation results show that the feedback teaching-learning-based optimization algorithm improves the precision and speed of the parameter identification in chaotic system markedly, so the feasibility and effectiveness of the algorithm are well demonstrated.