Real-time forecasting method of building energy load based on neural network elastic weight solidification

Abstract

The invention proposes a real-time prediction method of building energy load based on neural network elastic weight solidification, which includes two steps of offline training of neural network model and real-time fine-tuning of neural network model. The model offline training step includes six sub-steps, which are successively acquisition of building energy load historical data, data preprocessing, model input selection, model hyperparameter optimization, model parameter training, and parameter importance calculation. The real-time fine-tuning step of the model includes six sub-steps, which are real-time data acquisition of building energy load, data preprocessing, keeping model input consistent, keeping model hyperparameters consistent, fine-tuning the model using elastic weight curing, and updating parameter importance. In particular, fine-tuning of the model is carried out periodically in the form of a sliding window, so as to ensure that the model can adapt to changes in the law of building energy use. The method regularly fine-tunes the neural network model by using the elastic weight solidification technology, so as to realize accurate and reliable real-time prediction of building energy load.

Description

technical field [0001] The invention belongs to the field of building energy load forecasting, and relates to a neural network-based black box modeling technology and a continual learning technology based on elastic weight consolidation (elastic weight consolidation), in particular to a building based on neural network elastic weight consolidation A method for real-time forecasting of energy loads. Background technique [0002] Accurate building energy load forecasting is crucial for optimal operation and fault diagnosis of building electromechanical systems such as central air-conditioning systems and lighting systems. With the popularization of building automation systems, a large amount of building energy load data can be stored offline and acquired in real time, thus providing a reliable data source for building energy load forecasting technology based on neural networks. A large number of studies have shown that neural networks (BP neural network, recurrent neural netw...

AI Technical Summary

Problems solved by technology

This method can greatly reduce data storage costs and neural network training time, but due to limited training data, the neural network will not be able to learn the historical energy consumption laws of buildings outside the sliding window

Images