Spark-based demand side load prediction method

Abstract

The invention aims at combining a Spark big data platform and an LSTM neural network load prediction method, and avoiding the problems confronted by a single-machine data mining mode. In order to achieve the aim, the invention provides a Spark-based demand side load prediction method. According to the method, a load prediction model is constructed on the basis of an LSTM neural network method in the field of deep learning, so that power load can be accurately predicted; and parallel load prediction is carried out on the basis of Spark memory parallel calculation framework, the LSTM neural network algorithm is paralleled and parallel analysis is carried out on history load data, so that the load prediction time is shortened and the ability of processing big data by the LSTM neural network algorithm is improved.

Description

technical field [0001] The invention discloses a short-term load forecasting method based on a Spark parallel framework, which belongs to the technical field of power demand side and the technical field of big data. Background technique [0002] With the continuous progress of modern technology and the in-depth study of smart grid, the theory and technology of load forecasting have been greatly developed. At present, there are many mainstream methods applied to load forecasting on the power demand side, such as artificial neural network, support vector machine, autoregressive moving average model, etc., when faced with massive high-dimensional training data, they will face long training time and difficulties Local minima, easy overfitting and other issues. At the same time, power load is related to many hidden variables, such as light, wind, holidays, etc. These variables are generally difficult to obtain or quantify; it is also difficult to exclude noise data from massive ...

AI Technical Summary

Problems solved by technology

At present, there are many mainstream methods applied to load forecasting on the power demand side, such as artificial neural network, support vector machine, autoregressive moving average model, etc., when faced with massive high-dimensional training data, they will face long training time and difficulties Problems such as local minima and easy overfitting

At the same time, the power load is related to many hidden variables, such as light, wind, holidays, etc. These variables are generally difficult to obtain or quantify; it is also difficult to exclude noise data from massive data and extract effective information

With the development of smart grid, communication network technology and sensor technology, a large amount of basic energy consumption data has been accumulated on the demand side. Exponential growth, the traditional stand-alone data mining model has been difficult to meet the actual application requirements

Images