模式识别神经网络课件.ppt

State of charge (SOC) estimation of high power NI-MH rechargeable battery with artificial neural network (ANN) Speaker: Fu Weili 080305008 automatization ABSTRACT This paper presents a three-layer feed-forward back-propagation (BP) artificial neural network (ANN), whose output is battery state-of-charge (SOC), to estimate and predict SOC of high power Ni-MH rechargeable battery. Especially, the ANN can satisfyingly estimate SOC of battery whose starting SOC is not originally known after about ten minutes constant load discharging (CLD), and most of absolute values of absolute errors are not more than 5%. 1.Introduction of SOC SOC is a very important component of battery management system (BMS); SOC in this paper is defined as Eq. (1): 2. Concepts of high power NI-MH rechargeable battery The battery is able to discharge and charge at high rate current; Thirteen different constant current discharging (CCD) dates and two (0.15Ω/cell and 0.675 Ω/cell) CLD dates are obtained ; Six CCD data sets are selected to train ANN. 3. ANN architecture The first layer is input layer; The second is hidden layer, and its activation functions is logsig-moid functions which is defined as Eq. (2): LS(X) = 1/(1+ex) (2) The ANN output SOC is : SOC = W2* LS(W1* X + B1) + B2 where X is the input vector, B1and B2 are the bias vectors of ANN in the hidden layer and output layer. W1、W2 are weight matrices. 4. Selecting of ANN inputs Temperature factor is neglected; Based on the experience and knowledge of battery the following variables are initially selected as candidate inputs of the ANN: Discharging current i; Accumulated ampere hours Ah= ; Battery terminal voltage V; Time-average voltage tav(t) = ; Twice time-average voltage ttav(t) = ; ... The method to determinate input variables There are two basal parameters: i, v. Other parameters can be derived from the two and sampling time t.