Multi-Dimensional Time-series Mining And Its Application In EMS

With the new energy technological revolution growing, power grid intelligence is the focus of concern. It is the main emphasis to implement intelligent, economic, optimal dispatch and ensure information safety in open electronic information network.Because of the huge power grid, mass data management, information extraction and data mining follow. In this paper, we take advantage of data mining algorithm, such as support vector machine, timing sequence index, rule mining, to extract information from power grid data.In this paper, we do research into smart grid from following four aspects:(1)design of classification for smart grid data (2) analyze the relationship among factors according to association rule mining (3) short-term prediction on the basis of history data (4) power optimization for group.High-dimension time series monitor of smart grid can be views as classification in essence, so it require high classification accuracy and real-time. Traditional classification models are hard to adapt to this issue because of no real-time design in its mechanism. In this paper, we proposed a new incremental learning based smart grid real-time classification model. First we transform dynamic data into static data via sliding windows, and make sure that each data block consists of m records. Then we separate data into two parts, including tests and training. It is worthwhile to note that the raw data does not have label and we use k-means algorithm to mark labels.Smart grid data has many attributes. Each attribute establishes potential contact with each other. In this paper, we discrete smart grid data first, and evaluate how precipitation, wind direction, wind speed, sunshine and temperature affect load. Finally we provide weight of five factors so as to lay the foundation of prediction.Traditional predictions are usually based on artificial experience or rules, but effect is not ideal. In this paper we employ learning based prediction method, taking advantage of history data, extract key knowledge and produce prediction model.In term of power supply optimization, we need take consideration of many related factors. In this paper, we provide an iterative method, the main idea is to get rid of plans that are not suitable in space and finally we can get the optimum plan. This method can meet all the needs of each factor. Also we provide another simulated annealing method to build mathematical model, namely to get length of working hours effectively in limited time.