Nonlinear Forecasting Approach Of Heating Load

China is the biggest energy consumption developing country in the world. Especially, energy consumption for heating supply of north, northeast and northwest China in winter was account for about 27.2% of the whole society energy consumption. With large power consuming and low efficiency, heating supply is a key energy saving research field, which energy consumption occupied 3~4 times per square metre more than developed countries. In order to implement national policies of energy-saving and carbon emission depressed, it is very important to study the energy-saving in heating supply system.In recent years, through a constant exploratory work of domestic and foreign scholars along with experts in heating supply system operation and management, a series of effective forecasting methods have been developed. However, overview domestic and international research in present finds that a large number of methods get point forecasting results. Development of heating supply system puts forward the new request to heating load forecasting methods, but point forecasting methods can neither fulfill the needs of the energy saving control, heat dispatching, risk analysis nor reliability assessment in heating supply system. Therefore, it's of great theoretical and practical significance to analyze the heating load variation law and exploit the heating load interval forecasting method to realize the heating load probabilistic prediction.As we know, the key to carry out energy-saving is load forecasting and optimize the system configuration. In order to save energy, heat load forecasting methods were researched and forecasting software was developed in this paper. Load forecasting need to exploit characteristic of heat load from historical data and correlation factors. Hurst exponent is calculated to determine the nonlinearity in heat load time series, and largest Lyapunov exponent is computed to identify the presence of chaos in heat load time series. The simulation results illustrate that the heat load time series has the characteristics of nonlinearity and chaos.Support Vector Machine (SVM) is a novel nonlinear learning technique which is suitable for the nonlinearity in heat load. It has more superior functions than the traditional artificial neural network, which is based on the experience risk minimization principle. Therefore, the paper introduces this theory into the estimation of heat load forecasting, and mixes it with support vector machines to come up with the reconstruction of phase space and heat load time series crossover. Then Multi-Input and Single-Output Support Vector Regression (MISO-SVR) is applied to one-step horizontal and vertical heat load forecasting, and Multi-Input and Multi-Output Support Vector Regression (MIMO-SVR) is applied to daily heat load forecasting. Finally, the performances of above algorithms are analyzed by comparing the simulation results.Traditional forecasting methods are mostly point ones which is impossible to determine the range of heat load, in our opinions, for heating load forecasting problem, probabilistic forecasting is more objective and available for practical applications. Interval forecasting based on Support Vector Machine is presented to predict heating load in this paper. Then three interval forecasting approaches are proposed which based on SVR and error estimation, based on SVR and Markov Chains, based on Support Vector Intervals Regression (SVIR), respectively. Finally the forecast results are presented, and the simulation results of above three approaches illustrate that the forecasting confidence interval can provide a practical basis for heat dispatching, meanwhile, the average of confidence interval bounds can be defined as energy saving control setting value of heat supply substation.According to the characteristics of chaos presented in heat load time series, chaotic time series prediction methods have been introduced into the estimation of heat load forecasting to call for new methods and new ideas. Therefore, largest Lyapunov exponent and Volterra adaptive filter forecasting methods are applied to heating load forecasting, and the forecast results show that above forecasting methods are adopted to reveal the chaotic dynamics performance, so chaotic time series prediction methods are fit for heat load forecasting problem.On the basis of project application demands, heat load forecasting software is developed. Through general regulation, architecture of load forecasting software, framework of software functions and flow chart of load forecasting software are given, respectively. Then the software development environment, programming language, database and interface technique have been selected, respectively. Finally, the heat load forecasting software is developed with an engineering example.