Research And Application Of Short-term Load Forecasting Based On Electricity Market

The electric power industry is one of the basic industries,which are closely related to the economic and social development of the country.Short-term load forecasting plays a guiding role in power system security dispatch.Electricity market-oriented reform is the inevitable trend and requirement of the development of our country,in which electricity price market-oriented is one of the most important steps.Under the completely free operation mode of power market,the real-time changed price is an important factor affecting load fluctuation that increases the complexity of load forecasting.The accuracy and speed of load forecasting will directly affect the economic cost of operators in power market.Although China has not realized real-time power trading at present,it is of great significance to' study short-term load forecasting under real-time price with the assistances of experience and data from mature foreign markets.First of all,the historical load data are analyzed and show obvious regular characteristics in different time scales of year,month,week and day.These characteristics can provide references for accurate load forecasting.Then,a number of factors affecting load fluctuation are studied.The correlation between electricity price and current load and the correlation between historical load and current load are analyzed by using maximum information coefficient.Experiments demonstrate that these two factors have great impact on load and need to be considered in forecasting process.In view of the obvious time series characteristics of load,the LSTM network,which has been successful in dealing with time series problems,is chosen in this thesis as the prediction engine.Set the price,historical load and weather data as input vectors to build load forecasting model through two LSTM hidden layers and full connection layers.In order to highlight the impact of key factors on load and help the forecasting model to make more accurate judgment,the attention mechanism is introduced to give different weights to the input characteristics of LSTM network,and an Attention-LSTM network based load forecasting model is constructed.Taking real load data from Australia and Singapore as an example,the better prediction result and stronger robustness of the model in this thesis are verified through the comparison experiments with SVM,RNN and other models.Aiming at the improved application effect of load forecasting in production plan and cost-effectiveness,the probabilistic interval load forecasting model is also studied in this thesis.The historical load is partitioned by periods and load dynamics,and the load forecasting errors in each area are modeled through non-parametric kernel estimation method.Under given confidence,the deterministic forecasting results are transformed into probabilistic interval forecasting results.Experiments show that the method has high interval coverage and satisfactory interval prediction results.