abstract |
The invention discloses a time series prediction method for natural gas daily load, which comprises the following steps: analyzing the time series dependency relationship of historical natural gas daily load data; using the sliding window principle, determining the window length according to the dependency relationship, and extracting the time series within the time window Features; use heuristic search to determine the values of hyperparameters in the model, and use Adam optimization algorithm to obtain the values of parameters in the model; analyze the prediction error within the sample and perform white noise test; finally use the obtained model to predict the next weather day load. Aiming at the situation that the date affects the load value, the relationship between the date code mining and the natural gas load data is used to save the trouble of introducing exogenous variables and its own accuracy. At the same time, the heuristic search is used to determine the hyperparameters of the model, which increases the anti-interference performance of the model. , while the time window and recurrent neural network can fully exploit the characteristics of the time series itself, and ultimately improve the prediction accuracy of the model. |