Time Series Analysis In The Management And Prediction Of Residents Gas Consumption

The enterprises buy liquefied petroleum gas to user use it,there is a time interval. It is necessary to maintain a certain amount of gas.Reserves of liquefied petroleum gas needs equipment, person management, and there are security risks. At present, China mainly relies on imports of crude oil, but oil-producing countries political always instability, so oil prices often fluctuate.In addition, there is a large number of users of mechanical watches, for this part of the gas users, it need the gas staff enter user home check mechanical watches read-ing,recording into system,and withholding bank fees. This mode of operation have a low rate of meter reading and not quickly return of gas funds, Statistics show that the success rate of meter reading of enter home was only 30%, the success rate of charge also have 80%, so The causes of late fees up to 600 million per month(In a city as an example).The quality of the cash flow situation is crucial for business and development.the area of liquefied gas stock success forecasting can effectively reduce inventory and the capital investment,to avoid the risk of funds due to external risks and the existence of production safety hazards; the effective forecasting of individual residents gas consumption can provide decision support for the enterprise withholding, return the funds in advance, enhance the en-terprise’s cash flow situation.In order to effectively predict stocks, the paper obtained and analyzed 4497 resident us-ers’actual gas consumption in 2008-2013, and preprocessing gas consumption data, this pa-per propose use SARIMA model, exponential smoothing, BP neural network and wavelet neural network modeling of the total area of gas consumption,compare the effect of each model,get the best model to predict the next stage of the necessary area stocks.For individu-al residential users of gas consumption forecast, and this paper proposes use K-means clus-tering algorithm, combined with DTW distance as the sequence similarity measure for gas residential users modeling, "clusters" to form a "virtual area" and then use SARIMA model modeling virtual area total gas,get the forecasting model for individual users of gas con-sumption, forcasting individual users of gas consumption.Experiment results show that establishment of Area inventory model can effectively predict the next stage area necessary LPG inventories, the accuracy rate of 94%, compared with exponential smoothing, BP neural network, wavelet neural network also have a higher accuracy rate, it provide a reference for enterprise arrange inventories, and avoid the risk. establish a "virtual area" gas consumption model can effectively predict individual residen-tial users of LPG gas consumption, the accuracy rate of 70%, It can help the enterprises conservative withholding for reference.