| With the development of times change, the current information technology as a core of new military revolution, and already incoming roll efficiency.based on global logistical oil refinement security puts forward the new higher requirements. Strengthen military fuel consumption forecast, as one of the important contents for oil unit prepare application plan, fuel management department, decision-making authority distribution of oil index order, allocate the plan formulated has important auxiliary decision significance.This paper establish in lanzhou military fuel cost forecast model in the practical problems, compared the current several common prediction method were introduced, based on artificial neural network, especially the BP neural network to forecast, and the principle and characteristic of distinguished from other methods unique. Theoretically, the BP neural network for proof of this kind of like military fuel cost of complex non-linear system has a good approximation ability.Due to the standard BP neural network algorithm such as easy to exist in a local minimum value, slow convergence speed and forecasting precision are not high, so we need to get to the improvement of the prediction effect more ideal. According to different, mainly based on principle based on the gradient descending method and numerical optimization based on the two kinds of improving learning algorithm. This paper puts forward a case, namely the momentum as a representative adaptive learning algorithm and L-M learning algorithm as the main algorithm next forecast.Based on BP neural network, establish the prediction step in lanzhou, prediction model military fuel cost by using two kinds of learning algorithm toolbox of MATLAB, through networks are trained, and using the trained network respectively for military fuel cost forecast. Through the experiment, the simulation results of the comparison and analysis, it is concluded that the two types of learning algorithm by using numerical optimization learning algorithm is best prediction effect. |
