A deep learning method for probabilistic prediction of residents' load considering micro-meteorology and user patterns

Abstract

The invention discloses a resident load probability prediction deep learning method considering microclimate and a user mode, and the method comprises the steps: firstly collecting historical user power load and numerical value microclimate forecast and other related data to construct a two-dimensional multichannel feature map, and enabling the two-dimensional multichannel feature map to serve asthe input of a deep learning model; secondly, building a deep extrusion-excitation residual convolutional neural network model, and carrying out preliminary prediction of user electrical load probability prediction on the residential electrical load; then, based on a sparse-redundant characteristic representation method, extracting a characteristic mode in a daily load curve of the user, and performing uncertainty correction on a probability prediction interval; and finally, performing error analysis on a day-ahead resident load probability prediction result. According to the method, the microclimate data and the power utilization mode are combined to construct a new sample as model input, and the weather forecast data of a large number of microclimate stations near the area where residents are located are effectively combined, so that high-precision day-ahead user power utilization load prediction is realized.

Description

technical field [0001] The invention belongs to the big data analysis technology of electric power system, and specifically relates to a deep learning method of residents' load probability prediction considering micro-climate and user mode. Background technique [0002] The purpose of load forecasting is to predict the power load demand in advance and provide valuable guidance for power grid dispatching and power market planning. Reliable and accurate prediction results help to fully utilize power supply equipment and reduce energy consumption. According to different forecasting objects, load forecasting can be classified into different types, including system load forecasting, distribution network load forecasting, and residential load forecasting. Among them, compared with other levels of load, residential load usually has higher randomness and volatility, so its prediction uncertainty is stronger, and its prediction accuracy is more difficult to improve. In addition, re...

AI Technical Summary

Problems solved by technology

For traditional machine learning prediction methods, they usually have two common disadvantages: on the one hand, they are all trained based on the entire training data set in order to obtain optimal results under certain performance standards; but in this case , their training time can increase dramatically when faced with large datasets

However, it can be seen from the current research status that in the field of user load forecasting, how to build an effective and accurate deep learning model is still a major problem; in addition, the number of researches on constructing deep learning models to predict input is also very limited

Images