| The data management system of electric load is a systematic component of technical support of the electric marketing, implement the management effective technological means of demand side. It is gathered that it is the information taking application technology of the computer , modern communication technology , automatic control technology of electricity as foundation, deals with and real-time monitoring system. Power load forecasting has an important assistance function to the electric power system movement. Its precision will influence the economic and secure operation of power systems and quality of power supply. This thesis mainly aims at the short-term load forecasting, It will develop , predict partly to the theoretical foundation of electric load , concept the method does some to introduce, carry on the theory to analyze and research to the data management system of electric load at the same time .Using general decision support objects (DSO) of Microsoft Corporation a data mining model with the form of decision tree is designed and a daily load forecasting system is implemented according to the weather-load database of regional power network. After describing the DSO hierarchy structure, the constructing process of decision-tree data mining models for daily load forecasting is analyzed and the programming way for this model is given, furthermore, the load forecasting process by decision tree algorithm is implemented. The results of actual application and statistic analysis show that the presented system is intelligent, adaptive, versatile, reliable and accurate, it possesses the features such as self-study and full automatic load forecasting, therefore as an easy and practical load forecasting tool, this system is worth wide-spreading. Data mining methods will be used for data mining and forecasting model of the algorithm to extract the formation of a separate model of the algorithm, so that the combination of models and algorithms have greater flexibility.Using neural network prediction,the input variables decide the structure of neural network,this method proposed that may cause from historical sample knowledge data becomes finally to the forecast model modeling process the simple perspicuity,is advantageous for practical application. Using MATLAB establishment model, the 24 points for load forecasting, using multi-input and single-output neural network to forecast the daily value of the entire load. Because electricity load changes with environmental factors, in the input and output vector design join a weather characteristic a value in the importation. The basis the input and output vector design to the BP network. Finally, I import the 2000 Shanghai daily power load data into MATLAB, then short-term load forecasting simulation, simulation results show its better forecasting accuracy. The network model is smaller, short training time, and the advantage to consider different hours load difference with high precision. Prediction error below 15%. Proposed uses the different starting value to the power initialization method, trains many times to the network, in the certain extent overcomes the traditional algorithm convergence rate to be slow, easy to fall into the partial product small shortcoming.Two kind of algorithms, in view of forecast in the short-term date load diagram that uses the neural network the data processing method, is one kind of quite effective forecast plan, It easy to unify each kind of influencing factor, the network model structure is young, the curacy error is relatively small, the easy to operate, easy to promote, but the shortcoming is the model training time is excessively long; The decision tree algorithm model may very good withdraw the effective information from the magnanimous historical data, but because the daily user uses electricity the behavior as well as the influencing factor (weather, therefore temperature and so on) has too many uncertainty and unpredictability regarding the month, the season load forecast has the good forecasting result?... |
PDF Full Text Request