Research On Electricity-Energy-Control System Based On Short-term Load Forecasting

Considering the situation that electrical cost of industry at home and abroad includes two parts (basic cost and actual cost), an energy-saving electricity-consumption-control system is proposed for metallurgical enterprises with many high-power electric arc furnaces, which consume a great deal of energy, to reduce their basic electrical cost by evenly regulating the power and decreasing the maximum load. Efficient short-term load forecasting algorithm is the core issue of this study.Following achievements have been obtained:Detailed analysis to current similar technology at home and abroad is carried out. The main short-term load forecasting algorithms and their principles, methods and characteristics are discussed in depth. Actual background, necessity and significance of this study are explained.Considering the difficulties of selecting the electric power capacity of high-power electric arc furnaces resulted by drastic power fluctuation, a new arithmetic based on the threshold theory is presented originally. The energy function of threshold value is constructed by the method of variance analysis, and the base for selecting threshold value is obtained. The formulas for calculating the crossing intensity are deducted. The rationality of the selected threshold is verified by the crossing intensity of the instant power of power-supply system to the threshold.Based on historical load data, the method is put forward for determining the optimal data length of the grey model GM(1,1), and the forecasting results are amended by novel methods such as residual revision and filling innovation in proper order. The G-G-NN algorithm is proposed by combining the characteristics of grey theory, reconstruction-phase-space GP algorithm and artificial neural network (NN). Using grey theory and G.P algorithm, this new algorithm convert the original time series into the time series phase space with strong orderliness, and then the load is predicted by NN. The predicted results are of higher precision and better real-time than neural network.Aiming at the fluctuation and periodicity of the load series, the new method based on the wavelet excellent characteristics to analysis time and frequency is suggested, with which the signal mixed with different frequencies is decomposed into signals in different frequency bands, and different neural networks are used to forecast the data in different scale space, then the forecasting results are obtained by reconstructed. Different wavelet functions are compared and discussed and it is proved by actual example that the more accurate results are forecasted by this method.Another short-term load forecasting algorithm combining data mining algorithm and support-vector-machine method is provided. Firstly, the original data are preliminarily operated using clustering algorithm of data mining, the mass input data being compressed. Secondly, the clustering centre is chosen as the input characteristics of the support-vector-machine model, and the optimum core function is selected by cross-validating discrimination. Finally, the short-term load forecasting is finished. Actual examples show that the influences of the limitation and integrality of historical data and the complexity of factors on the forecasted results are minimized. This method has higher practical application value.According these achievements, the electricity-consumption-control software for metallurgical enterprises is designed and debugged by using Visual C++, then form the visual human-computer interactive interface, the forecasting, controlling and comprehensive management of the power load for steel-making plants being realized.