Energy consumption prediction method and device, computer equipment and storage medium

Abstract

The embodiment of the invention belongs to the field of research and development, and relates to an energy consumption prediction method and device, computer equipment and a storage medium. The energyconsumption prediction method comprises the steps: obtaining a timestamp, consumption data, basic data and external temperature data of a to-be-tested object, wherein the consumption data and the external temperature data respectively correspond to the timestamp; disassembling the timestamp to obtain a plurality of cyclic parameters; and inputting the consumption data, the basic data, the external temperature data and the plurality of cyclic parameters into a prediction model for prediction to obtain a prediction result. According to the energy consumption prediction method, the prediction result accuracy can be improved by reducing overfitting of the prediction result.

Description

Technical field [0001] This application relates to the field of research and development technology, and in particular to energy consumption prediction methods, devices, computer equipment and storage media. Background technique [0002] The existing prediction models in the industry often require the accumulation of sufficient historical data as the training set for model learning. However, to obtain a large amount of historical data, this process is more difficult and consumes more resources. Generally, the larger the historical data set, the prediction result The more accurate it will be; and all the objects to be measured need to be predicted together, then the historical data between the objects to be measured will affect each other, which will eventually cause a large deviation in the prediction results. However, in the actual application process, it often appears that the model does not have sufficient historical data when it is started early, that is, there is a cold star...

AI Technical Summary

Problems solved by technology

[0002] Existing forecasting models in the industry often require the accumulation of sufficient historical data as a training set for model learning. However, obtaining a large amount of historical data is a relatively difficult process and consumes a lot of resources. Generally, the larger the historical data set, the lower the prediction results. will be more accurate; and all the objects to be measured need to be predicted together, then the historical data between the objects to be measured will affect each other, which will eventually cause large deviations in the prediction results

However, in the actual application process, it often occurs that the model does not have sufficient historical data when it is started early, that is, there is a cold start problem.

In addition, the existing technology is to input the historical data of all objects to be measured into the model for unified prediction after obtaining each historical data, and the prediction result of each object to be measured is often affected by the historical data of other objects to be measured. Influence, increase the cost of obtaining historical data, and also cause unnecessary bias to the prediction results of each object under test

It can be seen that the existing technology has the problems of high cost of data acquisition and low accuracy of prediction results

Images