Study On Theory And Key Technology Of Photovoltaic System Operation

With the succesive development of the world economy and the increase of thepopulation size, the demand on energy is fast growing. The fossil fuels, however,isnow scaring daily, which will further worsen the environment. To deal with suchproblems, the renewable energy technologies have been attached more attentions inrecent years. The solar energy, one of the renewable energy resources, has greatdevelopment potential because it is a kind of unlimited resource, non-polluting andsustainable. Due to the instability of the sun light and the continues increase of powergeneration, there are some operational problems in the running process of aphotovoltaic system, such as the maximum power point tracking, power prediction,PV module temperature prediction and fault localization. These problems will befocused on in this thesis and the following five contents are studied.First, the solar light intensity and generating capacity of photovoltaic powerstations in Xuzhou, Jiangsu Province were counted and analyzed. Based on theobtained historical data of the photovoltaic power plant provided by Ed Solar EnergyTechnology Co., Ltd in Jiangsu, we thoroughly studied the monthly light intensity,temperature changes and the generation law of the photovoltaic power plant, and alsogave the corresponding charts and analytical results. These results provide somesupport on data for the building and operation of photovoltaic power plants in Xuzhouarea. The differences between the components temperature and the ambient one arefurther addressed. The corresponding results are expected to be bases of the followingstudies of those important problems existing in the running process of a photovoltaicsystem.In order to more accurately tracking the maximum power point of a photovoltaicsystem, an immune genetic algorithm based simulation model which can search itsmaximum power point is established here. For a given light intensity and temperature,the theoretical maximum power point of the current condition can be quicklycalculated with the simulation model, which can be adopted as a reference forvalidating the subsequent maximum power point tracking algorithm. Then, bycombining the advantages of constant voltage method and disturbance observermethod, we propose a novel maximum power point tracking algorithm. According tothe approximate voltage of maximum power point and photovoltaic system operatingvoltage, the disturbance steps are changed, which can well solve the contradiction of the tracking speed and the power oscillation of original disturbance observer method.Meanwhile, the problem of constant voltage method that the tracked maximum powerpoint may not a true one can also be refrained from. The effectiveness of the proposedmethod is experimentally validated.The output power of PV arrays is difficult to be predicted due to unpredictablechanges in sun light intensity. To solve this problem, the method of short-termforecasts of the output power is here addressed by introducing weighted supportvector machine and similar days. The weights of the training data for training thepredictive model are determined according to the similar degree of the selectedsimilar days. To demonstrate the performance of the proposed algorithm, we appliedthe obtained data to experimentally validate it. The effectiveness of our algorithm isillustrated and it outperforms the often used neural network in high predictionaccuracy with small number of training data.In existing research involving component temperature, its value is oftensubstituted with the ambient temperature for simplicity. But the two temperatures arequite different, which will greatly influence the accuracy of simulation and maximumpower point tracking in the constant voltage method. To solve these problems, wehere first focused on the prediction of the PV array components temperature by use ofGaussian Process. The prediction model based on Gaussian Process is constructedwith a set of samples. To sufficiently illustrate the performance of the algorithm,different sizes of samples are considered in the simulations. The results comparedwith the neural network based algorithm show that our method by use of GaussianProcess has a better prediction to a small sample of data.During the running process of a photovoltaic system, faults often occur atdifferent time and places, which will inevitably impact the output power of the system.Therefore, the accurate and reliable localization of the faults are very important. Tothis end, a fault localization method using Gaussian Process is then given here.According to the placement of the voltage and current sensors, the fault modesexpreesed by binary numbers are converted into integers and the Gaussian Process isconstructed to approximate these integers. With newly measured voltages and currents,the corresponding integer values of the system can be easily calculated and thecorresponding operating models can be obtained. A plenty of experiments e.g.,different fault modes, are conducted to validate the effectiveness of the algorithm, andthe results demonstrate that the proposed algorithm has a higher localization accuracy than the neural network one.