Research On Power Load Forecasting Technology Based On Event Knowledge Graph

Power load regulation and control is the goal that the power department continues to pursue and improve,and it is also one of the key indicators for the infrastructure construction of smart cities.If we do not have a good grasp of the overall power consumption situation and can not adjust the power supply policy in a timely and reasonable manner,it will cause a series of adverse consequences,such as various livelihood problems caused by the power cut in Northeast China in 2021.However,the traditional power forecasting methods are unable to distinguish the specific power consumption scenes,and when faced with multivariable power time series data,it is impossible to choose the variables pertinently,which affects the prediction performance and then affects the power load regulation.Based on this,the research of power load forecasting technology based on event knowledge graph is carried out in this paper.The main work is as follows:Firstly,the cause of power load change is complex,the existing power load forecasting methods can not distinguish between real-time power consumption scenarios,and can not select appropriate exogenous variables according to specific power consumption scenarios,resulting in the result that the accuracy of multivariable power load forecasting is too low.In order to distinguish real-time power consumption scenarios and select appropriate exogenous variables according to specific power consumption scenarios,a reasoning method based on event knowledge graph is proposed in this paper.The reasoning method is divided into three steps: the activation of the event knowledge graph,the reasoning and the selection of exogenous variables.The activation of the event knowledge graph is to distinguish the specific power consumption scene.The reasoning of the event knowledge graph is to find the other possible event sets that may lead to the occurrence of the current event.The selection of exogenous variables is based on the mapping table between the possible event set and the exogenous variables to find the relevant variables involved in the event set.In a word,the reasoning method based on event knowledge graph is used to distinguish the specific power consumption scene,and the possible related exogenous variables are selected.Secondly,the existing power load forecasting models do not distinguish the short-term and long-term time series in the time dimension,and the ability to capture the dependence on the input time dimension is limited.Aiming at the multivariable power time series data constructed by relevant exogenous variables and power load variables,a multivariable short-term load forecasting model MNLF based on memory network is proposed in this paper.First of all,in view of the different effects of the data of different historical periods on the predicted future values,this paper divides the long-term time series and short-term time series.In addition,using the structure of memory network for reference,the coding representation of long-term time series stored by memory components is set,and the weighted representation of long-term time series is obtained by calculating the correlation coefficient with the coding representation of short-term time series.and spliced with the coding representation of the short-term time series,so that the history window of the model is longer and the prediction performance is stronger.Finally,in this paper,the prediction performance of the model and the effectiveness of the reasoning system based on event knowledge graph are verified.First of all,the prediction performance of the model itself is tested and compared with the basic model,and the performance is improved.Then the ablation experiment is carried out on the part of reasoning based on event knowledge graph,and the performance of the model is compared before and after variable selection,which shows the effectiveness of the prediction model proposed in this paper.Finally,taking the short-term load forecasting model proposed in this paper as the background engine,a power load forecasting system is designed based on Spring Boot framework technology,and the forecasting results display and data display modules are designed to visualize the forecasting results of the model.