Data Mining Based Demand Analysis And Prediction Of Mid-long Term Electricity Market

As everyone knows, electricity is one of the most important basic industrial sources for economic development, which relates to the lifeblood of the national economy, social stability and plays an extremely important role in improving people’s living standard. To some extent, electric power industry has become a measurement index of a country’s development both in material and spiritual civilization. While, different from the disposable energy such as oil, coal and so on, power can’t be stored largely, which is determined by the nature of electricity. So research for the methods of electricity forecasting is of great significance to the safe and stable operation of the whole system.Currently, in our country, realizing the balance of power supply and demand is a great challenge faced by power grid managers, which will still exist for a long time. In some areas, insufficient power supply often appears because of falsely estimation of power demand especially in the peak season of power demand. While in other place, sometimes the demand of electricity market is estimated too high. Redundant power not only leads to the rising cost of power supply but also causes the waste of resources. Therefore, forecasting to electricity market is an effective way to realize the promise of effective supply of electricity.Compared with short term load forecasting, medium and long term load forecasting faces more challenges:Medium and long term load is affected by more external variables such as economics, climates, governmental activities, and so on, which don’t obey a certain distribution. What’s more, the data for medium and long term load forecasting is much less than that used for short term load forecasting. Sometimes, power grid managers need to make specific plan to power supply and charging standards based on the future demand of each industry. In this circumstance, it’s impossible to model for each industry because there are so many industries in a certain power grid.Based on the challenges mentioned above, this paper regards the provincial electricity market demand forecast as the research object. The methods of forecasting for medium and long term demand of provincial, urban and industrial electricity is researched, so as to solve the challenges mentioned above and improve the accuracy of prediction. This paper is arranged as follows:1. As mentioned above, provincial and urban electricity markets are influenced by a variety of external factors. So, in order to obtain satisfied forecasting results, these external factors should be considered detailed. Based on the principle of national econometrics, few indexes are selected as key influence index of power consumption. Then, intelligent forecasting model is used to fit the non-linear relationship between power consumption and key influence indexes.2. Industrial medium and long term load forecasting plays an important role in the daily management of power grid. But because there are vast of industries involved, it is impossible to build uniform prediction model for each industry. Therefore, in this paper, with the aid of clustering algorithm in data mining, industry power markets are categorized into several groups according to the similarity of wave characteristics of the time series. Then, prediction model can be built for each group, whose group members can be used to the training of model. So as to make the clustering group more suitable for forecasting model, this paper improves the traditional clustering algorithm. The modified clustering algorithm can produce groups whose members are with higher correlation to one another, thus the prediction model can match each industry with accuracy.3. In order to validate the effectiveness of the proposed framework for medium and long term electricity forecasting, this paper takes a certain provincial power grid of southern power grid as example. The precision of the forecasting model is tested. And the results demonstrate the applicability and the accuracy of prediction model, thus the framework of model are proved.