Search

Neural network is an important tool to solve the problem of nonlinear system prediction and control. It has been widely concerned by scholars. However, the existing neural network cannot adaptively allocate the weight of mixed kernel function according to the sample characteristics when it is applied to electric load forecasting. Aiming at this problem, short-term load forecasting algorithm based on adaptive fusion of mixed kernel function is proposed. Firstly, kernel functions are selected from the standard local kernel function and the global kernel function library to form a mixed kernel function. The weight variables and parameters of the kernel function are combined to form a new parameter state vector. Then a nonlinear parameter estimation model is established. Based on this model, the high-order cubature Kalman filter is used to estimate the parameter state, so that the local kernel function and the global kernel function can be adaptively fused. Moreover, the trained neural network is used to predict the load. Finally, the experimental analysis is given based on the actual grid data, and the effectiveness of the adaptive fusion of mixed function algorithm is proved.

This paper studies distributed secure fusion estimation problem in presence of eavesdroppers, where an eavesdropper can collect the messages from sensors to the fusion center through communication networks. To prevent eavesdropping, an insertion method of artificial noises is developed based on the fusion estimation error covariance matrix such that: i) The local/fusion estimation error covariance of the eavesdropper is unbounded; ii) The fusion estimation error covariance of the defender is bounded. Moreover, a sufficient condition is derived to guarantee the effective noise insertion strategy. Simulations are given to show the effectiveness of the proposed methods.

A novel filter design is presented for a class of multivariate nonlinear stochastic systems with uncertain dynamics. The system is subject to non-Gaussian process noises. The entire filter system is proved to be bounded in terms of a specified performance index, provided that the weighting function satisfies a certain condition. An analytical solution of optimal filter gain matrix is also given. Further, the estimation error system is exponentially and ultimately bounded in the mean square sense, even with the presence of system uncertainties. Finally, a simulation example is demonstrated to validate the proposed scheme.