Short-term Electricity Price Forecasting Solutions Based On Multi-models

At present, the electricity industry in the world-wide is undergoing profound changes. The reform of the power industry is to improve the efficiency of electricity production, rationalize the tariff formation mechanism, provide high-quality, safe power products, and promote the power industry itself healthy development and finally, to reform the entire society for the better economic and social efficiency.The reform of the electricity market is the current world trend of the development of electric power industry in power and international scientific research and engineering practice hot spots, while the most important and most critical part is price determination in the electricity market. Price is not only the electricity market supply and demand signals, but also control of the electricity market transactions as an economic lever. Therefore, how to determine the corresponding tariff reasonably according to the market demand directly affects the normal electricity market operation. How the electricity market in accordance with the relevant historical data to forecast the future market clearing price accurately is very significant for participants in the market.The load of the power system is a time-series, as well as the price, and theoretically, all the methods for load forecasting can be used for electricity price forecast, e.g., time series, artificial neural network, markov chain, wavelet analysis, combined model and so on. However, due to the price characteristics of the tendency, seasonality, heteroscedaticity, and other inherent, it makes the forecast electricity price forecasting much more difficult than load forecasting methods. And the current methods are not satisfactory independently.1. Through further study of the neural network and time-series, based on the characteristics of short-term price, we proposed multi-model short-term price forecasting methods with combination of time sequence and the neural network. These methods created short-term time-series price forecasting models, and modeled and analyzed PJM electricity market data, and then took the result of time series model forecasts as the input signal of the neural network for training. Through the examples of analysis on PJM electricity market, the composition of the model predicted well, which have greatly enhanced the accuracy of short-term price forecasts. The methods have been validated to be efficiency with the model and thus have a good prospect.2. Based on the characteristics of electricity, with the idea of the period-decoupled, we proposed period-decoupled based short-term time series price forecasting method. The main principle is as follows: analyze the sequence of electricity in different time, and establish different time series model respectively. Through the examples of analysis on PJM electricity market, the composition of the model predicted well, which have greatly enhanced the accuracy of short-term price forecasts. The methods have been validated to be efficiency with the model and thus have a good prospect.3. Based on the difference between power price and load, we took the average absolute percent error MAPE as a model of evaluation indicators, which can better reflect the accuracy of the forecasted results. The effective of MAPE are well verified by analyzing above two models. Finally, the paper gave a summary and outlook, and briefly introduced the short-term price forecasts and some of the difficulties facing the Institute and in-depth study to be done in the direction of a brief introduction.