Detection, mining and forecasting of impact load in power load forecasting

https://doi.org/10.1016/j.amc.2004.08.030Get rights and content

Abstract

People often try to smooth or eliminate impact load all together in traditional power load forecasting. This, however, could result in the loss of important hidden information. In other words, the impact load itself may be of particular important and they cause electric network collapse very easily. So there is a beforehand estimate to change and characteristic of power load, especially in impact load estimate, is the precondition that electric system carry through economy dispatching, reduce production cost and prevent electric network widespread blackout or collapse. This paper proposes a new method for electric deviate load and impact load detection, mining and forecasting. It solved the “Masking effect” and the problem on which the deviate data detection cannot be measured in deviate data analysis.

Introduction

Load forecasting is one of the important tasks in a utility. Accuracy of the load forecasting has influence on security and efficiency in electric. The main resource for any power generating utility is the generating units. Efficient management of these units means running them at minimum cost to satisfy the requirements of consumers. To achieve this purpose, starting-up and shutting-down the generating units should be performed according to schedule. The scheduling process is also known as unit commitment. If a power generating system is able to meet consumers, demand at both normal and emergency conditions the system is said to be secure. If this is not the case, the system is said to be insecure. The process of specifying whether a system is secure or insecure is called security assessment. The set of actions necessary to restore the secure state of a system is called security enhancement. Both security assessment and security enhancement need deviate load or impact load prediction. If the forecast is inaccurate the generation will be either above or below the required. The prediction is too high extra generating units will be put to operation without real need, the prediction is too low shortfall will take place, if low-frequency load-shedding sets fail to shed consumers load at the moment, it will cause widespread blackout or collapse on electric system. The kinds of events are not infrequence in the world. Correcting the second situation either by activating standby units or purchasing electric power is costly. Thus, in both situations the generating utility will pay extra cost. So deviate load or impact load detection, mining and study are very important tasks, but it is more difficult task too, and fewer studies in the world, the difficult stems from shortage of appropriate data, different constitute of customer loads and an understanding of the way they use electricity.

Deviate data detection has a long history [1] in statistics, and more recently, in data mining. However, most deviate data detections often with respect to some parametric model in time series [2] and focuses on a very small percentage of data objects, which are often ignored or discarded, so deviate data detection and mining is a difficult task. Deviate data detection and analysis can be categorical into three approaches: the statistical approach, the distance-based approach, and the deviation-based approach. The statistical approach and discordance tests are described in Barnett and Levis [3], but statistical approach often meets “masking effect” [4]. Distance-based deviate data detection is described in Knoor and Ng [5]. The sequential problem approach to deviation-based outlier detection was introduced in Arning et al. [6]. Sarawagi et al. [7] introduced a discovery-driven method for identifying exception in large multidimensional data using OLAP data cubes. Jagadish et al. [8] introduced an efficient method for mining deviants in time-series databases.

Section snippets

Load deviant data and impact load points are identified

Load deviant data detection and mining can be described as follows [9]:

Given a set of n load data points and k the expected number of load deviant data, find the top objects that are considerably dissimilar, exceptional, or inconsistent with respect to the remaining load data. The load deviant data mining problem can be viewed as two subproblems: (1) define what load data can be considered as inconsistent in a given load data set and (2) find an efficient method to mine the load deviant data so

Impact load forecasting

Grey system theory [10] has been extensively applied to various fields of data processing, modeling, control, prediction, system analysis, and decision making [11]. The grey system modeling techniques are characterized by

  • (a)

    no need of assumption in probability distribution of data,

  • (b)

    as small as 3–7 points of data required for model,

  • (c)

    little computational effort to constitute the model, and highly adaptive to system dynamic behavior.

In this paper, based on these characteristics, grey system model

Impact load identifying and forecasting

We carry through simulation analysis to the 96 points load of Lanzhou electric network and we see that Lanzhou power network has the similar load curves from Fig. 1.

Conclusion

In this paper, a novel approach for load deviate data and impact load points detection, mining and forecasting have been developed, which incorporates the Grey System Forecasting with deviate data analysis. To implement this approach, we statistically studies the impact load demand, including monthly and annual impact load, of the Lanzhou electric network. It has been shown that the proposed method can provide more accurate results than the conventional deviate data detection techniques.

References (11)

  • C. Chatfield

    The Analysis of Time Series

    (1984)
  • E. Knorr, R. Ng, Algorithms for mining distance based outliers in large databases, in: Proc. VLDB, New York, August...
  • V. Barnett et al.

    Outliers in Statistical Data

    (1994)
  • R.B. Murphy, On tests for outlying observations, unpublished Ph. D thesis, Princeton University,...
  • E. Knoor et al.

    Distance-based outliers: algorithms and applications

    The VLDB Journal

    (2000)
There are more references available in the full text version of this article.

Cited by (0)

1

The research of this author was supported by Natural Science Foundation of Guansu Province under Grant (ZS031-A25-010-G).

2

The research of this author was supported by the 973 Project of China (G19980306).

View full text