| Short term load forecasting is one of the most important tasks to ensure the efficient operation of the power system,which plays a key role in the stability,economy and safe operation of the power system.For the accuracy of load forecasting,it is of great significance in the power system,and it is the basis to ensure the reasonable dispatching of power system.The high-precision forecasting of power load is one of the key directions that scholars pay attention to.In view of this,based on the analysis of short-term load forecasting demand,this paper summarizes the principle and key technologies of BPNN forecasting algorithm,establishes a BPNN short-term load forecasting model considering the daily meteorological characteristics,and analyzes the prediction results and errors of the BPNN models with 10,20 and 30 hidden layers respectively.In addition,in order to improve the accuracy of BPNN model due to the initialization of weights and thresholds,GA is used to reconstruct the BPNN,and then a GA-BPNN network prediction model considering the characteristics of daily weather is established.According to the model,a district of Wuhu City in November is taken as the sample to train the model,and the daily load on November 21 is predicted and error analyzed.The main conclusions are as follows:(1)A short-term load forecasting model of BPNN power system considering the characteristics of daily weather is established.An example of short-term load forecasting of power system is analyzed by using BPNN.Through the application of BPNN model,the results show that the BPNN model has the best prediction performance,its absolute error is less than 2%,and with the increase of time.The trend of reduction can meet the requirement of 3% error in power load forecasting,and the effect is good.However,although BPNN model can achieve better prediction results in power load forecasting,there are some local over optimization problems.(2)The number of hidden layers has a significant impact on the short-term prediction of BPNN model in power load.The prediction effect of BPNN model with 20 hidden layers is the worst,but the calculation efficiency is the highest.When the calculation step is 5991,the target calculation error is achieved.The simulation effect of BPNN with 10 hidden layers is the best,with the best prediction performance,and the predicted value simulates the actual value better at the same time,the relative error and absolute error are minimum.3)In the modeling,the number of hidden layer nodes is compared and analyzed.It is found that the more hidden layer nodes in the BPNN,the better the prediction results.Increasing the number of hidden layer nodes in a certain range is conducive to reducing the training speed and reducing the length,but the prediction accuracy does not meet the requirements.When the number of hidden layers is 10,BPNN model has the best prediction performance,and the predicted value can better represent actual load with the the smallest relative error and absolute error,and with the increase of time and training step,the prediction results show a gradual improvement trend.The predicted value is more in line with the actual situation.(4)The BPNN is optimized by genetic algorithm,and then the GA-BPNN network prediction model considering the daily meteorological characteristics is established.In November,a district of Wuhu City was taken as the sample to train the model,and the daily load on November 21 was predicted and analyzed.The BPNN model optimized by GA effectively improved the accuracy of power load prediction of the neural network model.This improvement has achieved a very good prediction effect no matter from individual time or from the whole point of view,which shows that the GA algorithm is not only effective At the same time,it can effectively improve the application value of BPNN model in power load forecasting.The results confirmed.The prediction accuracy of neural network improved by genetic algorithm has been greatly improved. |
PDF Full Text Request