Big Data Technology And Its Application On The Analysis Of Power Plant Units

China’s energy industry is developing rapidly, and energy demand is still showing a growth trend. As the main energy supply mode, coal-fired power generation plays an important role in the energy structure, and it is the key to achieve the goal of energy-saving and emission-reduction. With the development of digitization and informationization technologies, the rise and rapid development of big data technology attracts attention from all fields. From the aspects of theory, method and application, etc., big data thought and technology will be introduced to electric power production, in order to promote China’s electricity production changing from the extensive development mode with high energy consumption, high emission and low efficiency to the green development mode with low power consumption, low emission and high efficiency. Modeling methods of power plant units based on big data technology are multidisciplinary and comprehensive technologies, which are suitable for the complicated structure characteristics of the current industrial production. It is of great significance to do multi-angle, deep-level and wide-range mining on power plant units’ data, and to promote the application of big data technology in the power plant units for improving the unit efficiency and energy utilization.The data characteristics of power plant units were analyzed. The definition of power plant units’big data, application mode and characteristics of big data modeling were summarized. Finally the theory, methods and ideas of big data modeling in power plant units were formed.Firstly, outlier detection methods of real-time data in the big data modeling process were studied. For multiple measuring point parameters in power plants, modified Grubbs criterion, with the median instead of the mean and with the weight coefficient introduced, was proposed to do outlier detection of multiple measuring point parameters. For single measuring point parameters, the modified Pauta criterion dependent on the related parameters was adopted. Through the verification of real-time operating data, it is showed that the method used can detect the abnormal points, and effectively eliminate data with gross errors in real-time data.Secondly, research on key feature parameter selection in the process of large data modeling was carried out. Mean impact value method was used as a feature parameter selection method, and multi-factor weight coefficient distribution method based on mean impact value were adopted. Quantile was introduced to modified mean impact value method, and the quantile impact value method was proposed to do feature parameter selection. And quantile impact value was used as multi-factor weight coefficient distribution method. The sensitivity analysis method based on support vector machine was used to select the feature parameters, at the same time the method of multi-factor weight coefficient distribution based on sensitivity coefficient was presented. Feature parameter selection methods above can eliminate irrelevant or redundant parameters to streamline feature parameters, which could effectively guarantee the accuracy of the model, reduce the model complexity, and reduce the modeling time.Newly, the main steam flow model was taken as the example to verify the validity of the feature parameter selection methods. From the viewpoint of big data technology, the relationship between industry analysis components and coal calorific value was excavated. Meanwhile, the off-line analysis model of coal calorific value based on industry analysis components was set up. In addition, through big data technology to excavate the relationship between online controllable operation parameters and coal calorific value, and using the sensitivity analysis based on support vector machine method to select key feature parameters, thereby on-line monitoring model of coal calorific value was established, which provides a solution for realizing coal calorific value on-line monitoring.Then, the modeling theory and methods of energy consumption characteristics in power plant units were carried out based on big data. SVR-based sensitivity analysis was put forward to analyze the feature parameters affecting energy consumption. SVR-based sensitivity analysis method to do feature parameter selection could directly solve the partial derivative, which could avoid the partial derivative being replaced by a small perturbation to approximately express the change of the energy caused by the change of feature parameters. Considering the correlation and dependence between feature parameters and energy consumption, the feature parameters which had great influence on the energy consumption were extracted as the input parameters of the model, and then the model of energy consumption characteristic analysis was established. At the same time, characteristics response of energy consumption under different loads, different samples and different feature parameters were respectively studied.Finally, research on evaluation index system and evaluation methods of state assessment, energy-saving and emission-reduction of power plant units was presented. On the basis of extensive research on the characterization parameters for unit state evaluation and energy-saving and emission-reduction, the two level calculation model of unit state evaluation based on information entropy-principal component analysis was established. Considering from two aspects of energy-saving （including coal-saving, fuel-saving, water-saving, electricity-saving） and emission-reduction （including NO_X, SO₂ and dust）, an evaluation index system of energy saving and emission reduction was proposed. Energy-saving and emission-reduction comprehensive evaluation model based on maximum entropy-projection pursuit principle was established, which can provide reliable theoretical guidance for units’ energy-saving and emission-reduction dispatching.