Data-Based Prediction And Optimization For Building Energy Consumption

With the scale development and application of green energy and renewable energy, the applications of distributed energy resources in the buildings (Group), which include shallow geothermal energy, solar energy and wind energy, etc., become a trend for the development of building energy supply system. However, the volatility and uncertainty of the real energy demand in building and the mismatch and unpredictability of hot/cold/electrical load demand leads to the imbalance between the energy use and the energy supply, which brings difficulty to the optimization operation of the building energy supply system, and limits optimized operation benefit of building environmental equipment and energy systems. Hence, the construction method of the building dynamic load forecasting model for the optimized dispatching of the building and energy system is studied in this paper, which include the following contributions:1.A dynamic simulation system of an existing building is built based on the mechanism analysis using the TRNSYS software. Because it is difficult to build an experimental platform for multi-system coupling buildings in a real application, the TRNSYS software is chosen as the platform for building energy consumption simulation in our research works. The simulation principle of TRNSYS software is discussed in this paper firstly, and the main parameters of building simulation are set according to an existing office building envelope parameters and actual operation parameters of the system. Then the experimental data, which can represent the relationship between the building load and its influence factor, are obtained by calculating the annual and hourly cold/hot/electrical load of the building.2. A data-based building load forecasting model is constructed. In order to obtain building dynamic load forecasting model with high precision and strong universality, the TRNSYS simulation data are used as the actual building operation data, and the key factors of the dynamic load of building is obtained by analyzing the data using correlation analysis method. Dynamic fuzzy C-means clustering algorithm is proposed to determine the node number of HCMAC neural networks, effectively reducing the number of the network nodes, and decreasing the computational burden of neural network parameter learning. Simulation results show that the maximum relative error of the building load forecasting model based on improved HCMAC neural networks is less than 3.9%.3. A least square support vector machine based on particle swarm optimization (PSO-LSSVM) is used to predict the meteorological parameters of buildings. Meteorological parameters are one of the main factors affecting the building load. A multi-input/multi-output (MIMO) weather forecasting model is built based on historical information, using the least squares support vector machine (LSSVM) as the model prediction algorithm and particle swarm optimization (PSO) to optimize the parameters of the LSSVM. The simulation results show that, the predict model can predict weather parameters in 140 minutes ahead, which including outdoor temperature and humidity.4. A distributed energy system optimal scheduling strategy based on dynamic load demand is presented. The optimization scheduling strategy for distributed energy system under knowing load demand is studied using the particle swarm optimization algorithm and the minimum energy consumption cost as the optimization objective, and a "need one then product one" optimization scheduling strategy is obtained. The simulation results show that the optimal scheduling strategy can solve the problem of the imbalance between the building energy supply system and the energy system..