关键词: 卷积神经网络, 双向长短期记忆网络, 综合能源系统, 负荷预测, 核密度估计, 随机森林

Abstract: Load forecasting in integrated energy systems faces challenges such as insufficient multi-source heterogeneous data fusion and inadequate uncertainty quantification mechanisms. This paper proposes a hybrid CNN-BiLSTM-RF-KDE model, which employs a Convolutional Neural Network to extract local features from load data, a Bidirectional Long Short-Term Memory network to capture bidirectional temporal dependencies, a Random Forest to handle high-dimensional nonlinear relationships, and Kernel Density Estimation to quantify prediction uncertainty, forming a comprehensive workflow. Additionally, an electricity-heat-gas multi-energy flow coupling model is constructed to analyze the impact of different carbon price intervals on scheduling strategies. Case studies demonstrate the high accuracy of the proposed model in electricity and heat load forecasting. On the training set, the coefficient of determination for electricity and heat load predictions reaches 0.93 and 0.96, respectively. On the test set, the values are 0.79 for electricity and 0.84 for heat. The predicted power generation or heat output of each device shows high consistency with the mean trend of equipment operation. By applying this model, more accurate load values closer to actual measurements can be obtained. These refined load values serve as fundamental data, enabling more reliable analysis and scheduling for integrated energy systems.

Key words: Convolutional Neural Network, Bidirectional Long Short-Term Memory, Integrated energy system, Load Forecasting, Kernel Density Estimation, Random Forest