Short Term Steel Plant Load Forecasting Based on Multi Sequence LSTM Recurrent Neural Network and Self‐Organizing Neural Network

ABSTRACT Due to the complex patterns, numerous influencing factors, strong temporal characteristics, and mostly nonlinear relationships of power load in steel enterprises, it has become a hot research topic in recent years. This paper proposes a nonlinear temporal method based on temporal features of long short‐term memory (LSTM) recurrent neural networks, which improves the accuracy of the LSTM method in forecasting steel plant load by extracting multiple related sequence temporal information. Firstly, a self‐organizing neural network is used to analyse the load curve of the steel plant, classify and mine the energy consumption patterns of the steel plant, and extract temporal differences between classes. Then, the Spearman correlation coefficient is applied to study the temporal correlation of steel plants under multiple time series, and intra‐class temporal similarity information is extracted. Research has found that there are multiple related time series in the data samples of steel plants, such as temporal differences between adjacent time points, the same time points of adjacent days, and different time periods of adjacent weeks, as well as temporal similarities within the same time period. And there is a strong dependency between these time series. Therefore, a steel plant load forecasting framework based on multi time series LSTM recurrent neural network is proposed. The framework was tested on a real dataset containing energy consumption data from a steel plant in a certain region of China, and its performance was superior to other load forecasting methods. The experimental results show that this method can effectively utilize multi‐sequence information and successfully capture the dependency relationships between these sequences.