The Research On Heating Load Forecasting For Thermal Stations Based On Time Series

Energy shortage is the most important issue of every country in the world that have toface, therefore, the rational use of energy to improve the efficiency of energy use is the key tosolving this problem. Located in the Northern Hemisphere, winter heating energyconsumption of northeast, northwest and north of China is huge, accounting for about27%ofChina’s energy consumption. The main problem is lower heating efficiency, and the heatingenergy consumption per unit area is2-3times than western developed countries. Buildingheating is one of the important areas of national implementation of energy saving. In thispaper, the heating stations load forecasting is the manifestation of the heating energy research.The key technology of the heating energy saving is load forecast research and implementationof the system to optimize the allocation of the heating station. Load forecast for heatingstation is to achieve the guarantee of quality heating and energy efficiency.The article studies heating load data prediction research based on time series analysis ofthe various methods. The first part is basis data acquisition of the heating station in theselected northeast region, raw data obtained by calculating the heating load, and calculatedload sequence data preprocessing in three steps. First of all, the load of abnormal datasequence vacancies and vacancies abnormal data is processed according to the the anomalydata processed method for processing. And then, the abnormal data sequence processed issmooth identified. Finally, because of time series analysis methods on the stabilityrequirements of the data, we need to take determined to be non-smooth load sequenceaccording to its non-stationary types of differential method to smooth processed of thesequence. The stationary sequence after processed is considered to load forecast study samplesequence.According to time series model for the heating load sequence to identify, select ARmodel in the time series as the heating load forecasting model lateral prediction andlongitudinal prediction. First F criteria determines the order of the AR model, then theYule-Walker identified AR model parameters to determine the prediction model. In order toimprove the accuracy of load forecasting, we need to increase weights of the least-squaresmethod the load sequence based on the cross on the basis of the transverse of forecasting andlongitudinal forecast.According to the trend and seasonal heating load existing, the paperdetermine the orderof the model by AIC (BIC) criteria, and then apply the maximum likelihood method to identify the model parameters of the product of seasonal ARIMA model forecast of loadsequence research. As the product of seasonal ARIMA forecasting method is not sensitive tothe load sequence mutation part, smooth part of the forecast accuracy is relative reducing.This paper applied Kalman back stepping method to improve multiplicative seasonal ARIMAheating load forecasting method to solve this problem.Finally, experimental simulations have been done of the cross forecasting methods andmultiplicative seasonal ARIMA forecasting methods respectively. Simulation is based on thesequence of samples collected including horizontal prediction, simulation of longitudinalprediction, cross-prediction, the product of seasonal ARIMA forecast the product of theimproved seasonal ARIMA forecasting method. Comparative analysis of the simulationresults of the various methods, and the engineering applications reference recommendationswere given.