The BP Neural Network Based On GA To The Forecasting Of Earthquake Emergency Supplies Demand

China is located in the Eurasian seismic belt and the Pacific seismic belt. Because of the earthquake disasters occurred frequently in recent years both hurt China's economic level, also threatened the life safety of people, In the present research level and technology level, the expert can't scientifically for accurate prediction of earthquake. Although the occurrence of earthquake is inevitable, it can improve the efficiency of rescue, so as to reduce the loss of the earthquake. After the earthquake, the people of earthquake disaster area urgently need a large number of emergency supplies. Because it can't collect immediately the needs information of emergency supplies, the supply of goods can't be determined, which would make the distribution of seismic materials is still very rough, resulting in a serious threat to the survival of people in disaster areas, while not conducive seismic disaster relief work. If we can forecast the demand of emergency supplies after the earthquake disaster which can not only guarantee the smooth progress of relief efforts, but also can provide some convenience for the government organization and scheduling, to promote the smooth implementation of the rescue work.In this paper, we study the demand forecasting of earthquake emergency supplies. By reading a large number of relevant literatures to compare the direct and indirect prediction model, we use the indirect method of prediction. First using genetic algorithm improved BP neural network algorithm to predict the number of casualties in the earthquake. Then use the number of casualties and the relationship between the different types of materials, to join the safety stock theory to predict the demand for emergency supplies.The author establishes the prediction model of the emergency material demand of earthquake with BP neural network by the improvement of genetic algorithm on the basis of the related thesis about the demand forecast of emergency supplies. Firstly, the author selects seven predictors consist of magnitude, earthquake intensity, hypocenter depth, fortification intensity, level of pre-alarm, occurrence time and population density. Because the damage of multiple small earthquakes is little and people can help themselves, so the author selects 25 cases of the earthquake data above the 6.0 magnitude. Secondly, the author uses principal component analysis technique to do with the raw data so that the variable is more representative. Because the BP neural network is easily lost into local minimum at the time of the prediction of the earthquake, so the author applies genetic algorithm to enhance the global searching ability. The author compares the BP neural network to that with the improvement of genetic algorithm through the training and testing by the implementation of Matlab so as to prove that the error of the latter is litter than the former as well as the fitting effect. Thirdly, it is known that we can get the number of the injuries population by the forecast of the casualty rate in the earthquake and then we can predict the demand for emergency supplies with the knowledge about the inventory management. Then it predicts the earthquake in Sichuan province Luhuo County the number of drinking water, compressed biscuits, anti-inflammatory drugs. It is not only the foundation of accurate distribution system, but also the support of rescue work in the aftermath of disasters. All in all, it is important to apply the prediction model into the distribution of emergency supplies to avoid the imbalance between supply and demand and delay of material supply in the earthquake.The innovation of this paper is to use the genetic algorithm to optimize the initial weights and thresholds of the BP neural network in the prediction of earthquake emergency material demand.In this paper, there are some shortcomings. In terms of variable selection, based on the data availability, timeliness and picked only seven variables. But the factors that affect the earthquake casualties are also more complex, therefore subject to continue to study. In establishing the structure of the neural network, the determination of some parameters is established by experience or try algorithm, it will makes the prediction error. Using genetic algorithm to improve the BP neural network's initial weights and threshold, the forecast effect is better than before. In order to make the model fitting ability stronger, it also need to further research the model and improved. In the demand for materials to estimate assumes that supplies enough under the condition of not considering if goods cannot meet the needs of all the points completely, we should consider to different point of the affected the demand for goods and to determine the demand for the urgent degree, the major influence on supplies the actual distribution work.