Data center station multivariate load prediction method based on hybrid model prediction

Abstract

The invention provides a data center station multivariate load prediction method based on hybrid model prediction, and relates to the technical field of automatic control. According to the method, multivariate data of a data center station are divided into spring and autumn, summer and winter scenes, multivariate load prediction is carried out on the data in the scenes, feature analysis and normalization are carried out on the multivariate load data by adopting a GRA method, the processed data are input into a QPSO-BP neural network for prediction, and in the aspect of a prediction algorithm, a QPSO-BP neural network and an XGBoost model are adopted for parallel prediction, deep learning and machine learning technologies are applied to load prediction at the same time, two integrated learning modes are effectively combined, the advantages of the two models are fully played, and the model which is more stable and higher in generalization ability can be obtained. The hybrid prediction model can actively enrich the characteristics of input data with single dimension, avoid the influence of data errors caused by human factors on the calculation precision in the data acquisition process, and also can realize high-precision load prediction under the special conditions of large load fluctuation and the like.

Description

technical field [0001] The invention relates to the technical field of automatic control, in particular to a multi-element load forecasting method for a data center station based on a mixed model forecast. Background technique [0002] In recent years, with the rapid development of Internet technology, the scale and number of data center stations have been rapidly expanded. According to statistics, the power consumption of data centers in my country has accounted for 1% of the total power consumption in the country, and the load of data center stations has become a body. considerable electrical load. Under the requirements of fast and accurate dispatching of the power system and system security and stability, the prediction accuracy of the data center station has become a top priority. [0003] The load of the data center station is mainly divided into two categories, one is the server load for processing data, and the other is the storage, lighting, cooling and power distri...

AI Technical Summary

Problems solved by technology

The traditional load forecasting method often selects one factor to perform a single mapping analysis on the load, ignoring the influence of other factors, and does not consider the linkage relationship between various influencing factors, which leads to inaccurate analysis of load characteristics The formulation of the electricity plan will have an impact, and the accuracy is low

In addition, traditional forecasting models such as time series models, neural network models, and artificial intelligence optimization models have their own advantages and disadvantages. The assumptions and calculations of time series models are simple and adaptable, but the extrapolation effect is poor and the prediction range is small; neural network models The fitting effect is good, and it has the ability to deal with nonlinear data, but the model is unstable and depends on the characteristics of the data; the artificial intelligence optimization model can be used in combination with other methods to improve the prediction accuracy, but it is easy to fall into local optimum

In addition, traditional forecasting algorithms have some typical limitations: the error is not sensitive to changes in weight values, the error gradient changes very little, the adjustment time is long, the number of iterations is large, and the convergence is slow. The output layer of the neural network is easily trapped in a local minimum. There are certain defects in terms of stability and stability, and the above problems pose challenges to the accurate load forecasting of the data center station

[0004] Traditional forecasting methods do not fully exploit the massive amount of dormant historical operating data, and often predict the load in a single scenario, ignoring the difference in load at the time level, and not considering the various factors in the system that will affect the forecasting accuracy of the system. Moreover, the adjustment time of the traditional forecasting model is long, the number of iterations is large, and the convergence is slow. The output layer of the neural network is easily trapped in a local minimum, which has certain defects in the accuracy and stability of the forecast. The above factors will lead to inaccurate system forecasts.

Images