Research Of Power Load Forecasting Method And Smart Campus Application Based On Machine Learning

In the context of building a strong smart grid,domestic universities have accelerated the construction and improvement of smart campus energy consumption management systems.The role of low-voltage side energy management has become increasingly prominent,especially the power load forecasting task,which provides strong data support for the overall power distribution and energy saving planning of colleges and universities.Aiming at the trend of quantization and multi-dimension of campus power data,this paper solves the bottleneck of prediction performance that traditional machine learning prediction algorithms will encounter large cumulative errors.Firstly,a short-term power load combination forecasting method based on snapshot feedback mechanism is proposed.Then,because the peak load prediction plays an important role in the optimal operation and safety and stability of the power grid,a method of peak load prediction based on the multi-time sequence collaborative model is proposed in this paper.Finally,the design and implementation of smart campus power load forecasting system.The main work of this paper is as follows:(1)A short-term power load combination forecasting method based on snapshot feedback is proposed.In view of the problem that VMD algorithm needs to set K value in advance,and the K value will greatly affect the poor effect of the prediction model and the large accumulated error of the LSTM model in multi-time load forecasting,a short-term load forecasting model(VMDSF-CSLSTM)with snapshot feedback mechanism and cycle sliding window strategy is proposed.Four open data-sets are used to compare the short-term load forecasting results of the model with other models,and the effectiveness of the integrated model is verified.Finally,VMDSF-CSLSTM integrated model is applied to the actual short-term load forecasting of four transformers in Xiangtan University.(2)The peak load forecasting method based on multi-time series collaborative model is designed.Firstly,a multi-time series collaborative peak load prediction model(ARIMA-CSLSTM)based on additive auto-regressive integration sliding average model combined with long and short term memory network with cyclic sliding window strategy is proposed,which provides a peak load prediction method combining internal and external timing.Then,four open data sets of power load are used to compare the results of the proposed model with other peak load forecasting models.Finally,ARIMA-CSLSTM is applied to the peak load prediction of four groups of transformers in Xiangtan University.(3)Design and implement the intelligent campus power load forecasting system.Firstly,the requirements analysis and system design of the system are described in detail,and the main functional module design and application interface of the system are described in detail.The system can carry out load characteristic analysis,short-term load prediction and peak load prediction.According to the requirements of building a strong smart grid,the general idea of the distributed client side and the front and rear end separation system is adopted to realize the intellectualization and visualization of the system.