Short-term Thermal Load Forecasting Based On Optimized BP Neural Network

Nowadays,the energy consumption of buildings is about one-third of the energy consumption of the whole society.It is the field with the greatest energy-saving potential and would certainly be one focus of energy-saving plans.Within this category,the central heating in northern cities accounts for 21% of building energy consumptions.As the management is too robust to target demand accurately right now,the centralized heating system would have to develop in a refined direction,which also puts higher requirements on the accuracy of short-term load forecasting.Based on the analysis of the influencing factors and characteristics of thermal load,as well as the importance and availability of each influencing factor,this paper determined the input variables and pretreatment methods of the load forecasting mode.Taking a heating system in Chengde as an example,the BP neural network was used to establish the thermal load prediction model,and a variety of improved learning algorithms were applied for comparison.The Beetle Antennae algorithm is introduced to optimize the BP neural network in terms of initial weights.The results are compared with those obtained by particle swarm optimization.Furthermore,the optimal particle swarm optimization algorithm is used to analyze the influence of historical disturbance on the thermal load prediction model.The paper draws the following conclusions:(1)In view of the fact that most projects cannot obtain solar radiation intensity,the solar radiation intensity can be replaced by the weather condition and the solar elevation angle as input variables of the load prediction model.(2)BP neural network load forecasting model based on historical heating data and outdoor meteorological parameters is established,and standard BP neural network and four optimization algorithms are used to calculate time,modeling stability and modeling accuracy.By comparative analysis,the momentum adaptive method is the most ideal learning algorithm.(3)Using the Beetle Antennae algorithm to optimize the initial weight of the momentum adaptive BP neural network can improve the calculation accuracy and enhance the stability of the model.Furthermore,the comparison results show that the stability of the particle swarm optimization algorithm is better than the Beetle Antennae algorithm.(4)By studying the influence of hysteresis on thermal load prediction,increasing the historical data of input parameters,and using the particle swarm optimization algorithm to establish a thermal load prediction model for momentum adaptive neural network,it is learned that the prediction performance of the model is better considering the hysteresis effect.