Market-clearing Price Forecasting In Day-ahead Market For New South Wales

The core issue of the electricity market reform is to establish a reasonable electricity price mechanism.As China's power production shifts from a planned model to a market model,how to forecast electricity prices in the future is the focus of attention.With the rapid development of big data and artificial intelligence,machine learning theory has become an emerging forecasting method.The electricity price method in Australia's electricity market is similar to the electricity price method after China's electricity reform.Therefore,this thesis uses the data of the Australian electricity market to study the effectiveness of machine learning theory in electricity price forecasting,in order to provide ideas for China to formulate reasonable electricity price in the future.This thesis divides the electricity price market in New South Wales,Australia,according to the season,studies the characteristics of electricity price in each season,and finally determines the demand as one of the input variables of the model.In order to establish a better prediction model,this thesis uses the method of random forest feature extraction to select the input variables of each season,which can eliminate the redundant information.Then five models of decision tree,random forest,support vector machine.BP neural network and extreme learning machine are established to forecast the electricity price of New South Wales.Because random forest is a forest composed of many decision trees randomly selected from training data.Then the forest is used to forecast and select the classification with the largest number of votes.The correct rate of collective prediction is higher than that of decision tree.So random forest can be regarded as an optimal decision tree.Compared with BP neural network model,the extreme learning machine model does not need to update the weight and threshold,only need to set the number of hidden layer neurons,so the extreme learning machine is an improved neural network.Support vector machine is the mainstream machine learning algorithm.Using these five models to predict electricity price,we can compare and comprehensively analyze the application of the machine learning algorithm in the field of electricity price forecasting.According to the results of single model electricity price forecasting,the performance of BP neural network and decision tree in the five models is relatively general.BP neural network has good generalization ability and can deal with non-linear problems such as electricity price,but its training time is long,and it is easy to get local minimum value on the training process,resulting in training failure.Decision tree is fast and easy to understand,but the prediction results are easy to over-fit.Random forest has a good effect in forecasting the electricity price on the peak value,and the introduction of randomness makes Random Forest have a good anti-noise ability.The generalization ability of support vector machine is poor,and the actual price curve fluctuates greatly.The results show that the prediction effect of support vector machine is not ideal.Extreme learning machine has the advantages of fast calculation speed,simple parameter adjustment and great research space in electricity price forecasting.Combining the forecasting results of all models,the accuracy of forecasting electricity price with random forest is obviously better than that of other models.Random forest and extreme learning machine model can achieve good results in seasonal electricity price forecasting in New South Wales,Australia.The forecasting effect of random forest in autumn and winter is better,and the forecasting effect of extreme learning machine in spring and summer is better.