SFCL: Electricity Consumption Forecasting of CNN-LSTM Based on Similar Filter

Electricity consumption forecasting plays an important role in engineering and daily life. In the practical application of electricity consumption forecasting, multi-step forecasting often brings more valuable information than single-step forecasting. According to the classification of forecast range, prediction algorithms can be divided into: long-term prediction, short-term prediction and ultra-short-term prediction. However, due to the nonlinearity, instability and randomness of electricity consumption data series, the prediction series is unstable. A MAPE based similar filter (SF) algorithm is proposed for multi-step short-term electricity consumption forecasting. The SF algorithm in this paper obtains a similar sequence from the similar historical electricity consumption sequence, which is more conducive to fitting the Neural Network. The SF algorithm is combined with the feature capture and trend learning of Convolution Neural Network and the Long Short-Term Memory neural network to form the CNN and LSTM algorithm based on SF (SFCL) to predict the electricity consumption sequence in multiple steps. Experiments show that SFCL has good prediction effect.