Residential Short-Term Load Forecasting Based on CNN-LSTM with Consumer Behavior pattern

Daily electrical power needs are dramatically rising around the world. To support the side of the Power Management System in residential distribution systems, Short-Term Load Forecasting is crucial. However, the resident's behavior is somewhat improvised and unplanned in advance, causing extreme fluctuations and irregular data patterns in predicting residential loads, this led to creating a significant challenge for accurately predicting residential loads. Recent studies have focused on several variables to improve prediction, such as weather data and previous loads, the influence of the population consumption behavior factor on predicting residential loads has received less attention despite being a key component impacting energy consumption patterns. This paper proposes using Deep Learning (DL) techniques to predict a residential load a hybrid model of Convolutional Neural Networks (CNN) and Long Short-Term Memory to enhance the accuracy of prediction for private data of a Libyan family, Human Consumption Behavior will be added as an effective factor, will be extracted from a survey of consumption behavior of a family in the Benghazi city. The error rate is measured with Mean Squared Error, Mean Absolute Error, and Coefficient of Determination $(\mathbf{R}^{\mathbf{2}})$ in this work. The experiments results showed the superiority of the model with noticeable improvement in the prediction results with using of the Consumption Behavior Factor (HCB) through decreasing the error metrics from (0.103 MSE, 0.222 MAE 0.898 $\mathbf{R}^{\mathbf{2}})$ into (0.057 MSE, 0.155 MAE, 0.944 $\mathbf{R}^{\mathbf{2}}$ ) comparing to the slight improvement of (LSTM) model with using of this factor from (0.549 MSE, 0.541 MAE, 0.347 $\mathbf{R}^{\mathbf{2}}$ ) into (0.426 MSE, 0.441 MAE, 0.523 $\mathbf{R}^{\mathbf{2}}$ ).