Multi-target deep belief network for mid-term power load predictions

Abstract

Provided is an issue of processing mid-term power load predictions by using multi-target deep belief network based on decomposition. The method comprises the following steps: firstly, decomposing databy adopting empirical mode decomposition and combining decomposed data and original data into a new data set; secondly, using multi-target data mining, serving a single DBN as a candidate solution ofa multi-target optimization algorithm, iteratively evolving multiple DBNs a under the driving of an MOEA / D algorithm at the same time, so that accuracy and diversity of a model are considered, usingcross validation for samples after circulation is finished, to prevent overfitting; then, proposing a two-stage strategy to screen the prediction model, and finally allocating a weight vector to a single prediction model by utilizing ensemble learning. According to the invention, parameters of the prediction model can be optimized and selected, and the efficiency of the algorithm and the prediction accuracy are effectively improved.

Description

technical field [0001] The invention relates to the field of medium-term power load demand forecasting. To be precise, it uses decomposition-based multi-objective optimization deep belief network to deal with the problem of power load demand. Background technique [0002] Power load forecasting takes power big data as the research object to carry out forecasting work. From the perspective of forecasting objects, power load demand forecasting mainly predicts future electricity demand, and improves the accuracy and efficiency of load demand forecasting to promote power companies to design reasonable power distribution plans, so as to improve the social effects of the power system and economy. [0003] Algorithms used in electric load demand forecasting play an important role in electric load forecasting. At present, the methods of load demand forecasting mainly include neural network method, regression analysis method, trend extrapolation method, elastic coefficient method, ...

AI Technical Summary

Problems solved by technology

This not only increases the workload, but also may cause overfitting of the training set and increase the error of the test results; therefore, a method is needed to enable the neural network to independently select parameters

Images