Multi-time-step heat supply gas consumption prediction model based on attention mechanism

Abstract

The invention relates to an AMS-LSTM prediction model, which is mainly technically characterized in that a heat supply gas consumption prediction model is provided, the model takes a time step as a hyper-parameter for model optimization, and the time step is optimized and selected based on a Bayesian hyper-parameter optimization algorithm. In order to further improve the prediction precision of the model, the AMS-LSTM prediction model introducing an attention mechanism is constructed according to the attention mechanism, and the accuracy of the model is verified through test set data. The model is reasonable in design, and in order to verify the accuracy of the model, heat supply data of a 10-minute time scale is adopted as a data set of an experiment. Prediction effects of the AMS-LSTM model and the LSTM model at different time steps are evaluated through two evaluation standards of RMSE and MAPE. The result shows that the model provided by the invention has a good prediction effect no matter in a 10-minute time scale or a 1-hour time scale. The invention has good applicability to gas consumption prediction in the heat supply field, and gas consumption prediction can be more effectively carried out, so that energy is effectively utilized.

Description

technical field [0001] The invention belongs to time series data prediction in the field of heat supply, and relates to a time series prediction model of multi-time step LSTM based on an attention mechanism. Background technique [0002] Now time series forecasting is one of the hot research directions. Time series is a set of data recorded in chronological order, which is commonly found in many fields such as transportation, finance, logistics, science and industry, such as changes in weather data, changes in stock price sequences, and changes in power loads. Changes in status and development laws occur over time. Time series data has certain regularity in the process of change. [0003] Time series data mining technology is to obtain the inherently unknown research object from a large number of time series data records, but has high application value and information with strong time characteristics, and predicts the trend of time series according to the needs of differen...

AI Technical Summary

Problems solved by technology

All the information of the input sequence needs to be stored in the encoding vector, and the encoding vector is used for effective decoding. For long sequences, the previous information transmission will be lost.

Images