| With the worldwide energy crisis and environmental degradation problems,the renewable energy share of total energy consumption is gradually increasing.Due to the advantages of low-carbon,environmental friendliness,resources richness,and adaptation to local conditions,distributed photovoltaics(PVs)play an important role in country energy development strategies around the world.PV systems can generally serve for 25 years.So,how to ensure their efficient and economical operation after construction has become a new hot spot in current research.However,in actual operation,due to the large scale,wide distribution,and low degree of intelligence of distributed PVs,it has caused great difficulties for traditional state evaluation and operation and maintenance(O&M),seriously reducing the efficiency of distributed PV power generation and affecting its economic benefits.To this end,based on the needs of distributed PV intelligent state evaluation and O&M,this thesis researches four key technologies: PV array fault diagnosis,PV module dust accumulation state evaluation,optimal O&M cycle determination,and large-scale distributed PVs O&M resource scheduling.In this way,this thesis provides theoretical and technical support for the construction of distributed PV state evaluation and O&M system in the future.The main innovations and work are reflected in:(1)Array is the most important part of the PV system.Most of the PV system faults are attributed to the PV array and the module.Therefore,it is necessary to study the online fault diagnosis technology of the PV array.This thesis first analyzes the output characteristics of the PV system and the characteristics of open circuit,short circuit,shading,and aging faults.Second,since the number of the PV fault samples is sparse and unbalanced with the number of normal samples,Generative Adversarial Networks(GAN)is introduced in this thesis to enhance the fault samples.On this basis,the traditional Na?ve Bayesian model is improved,and a distributed PV fault diagnosis method based on the fine-tuning Na?ve Bayesian model(FTNB)is proposed.This method collects the PV inverter maximum power point data and environmental data as input,does not require additional measurement equipment and is suitable for distributed PV scenarios.Finally,the effectiveness of the method in diagnosing 4individual and 2 hybrid faults with ideal and noisy data is verified by simulation.(2)Tiny particles in the air can easily accumulate on the surface of the PV module and form dust accumulation.Dust accumulation is an important factor in reducing PV system power generation.It is necessary to conduct a dust accumulation state online evaluation method of the distributed PVs to guide the O&M.PV systems are generally exposed to outdoor environments,and their output power is bound to decrease after long-term operation.First,to improve the accuracy of the evaluation,long-term historical operating data of the PV system are collected,and a method for calculating the annual degradation rate of the system based on field test data is proposed.Second,fuzzy C-means(FCM)clustering is used to cluster PV data into four types: sunny,cloudless,cloudy,and rainy according to 8 indicators.The Deep Neural Network(DNN)model based on the Levenberg-Marquard Back Propagation(LMBP)algorithm is used to fit the PV output of the cloudy weather,and the light GBM is used to fit the PV output of the remaining three weather types.Then,the fitting value is corrected according to the PV operation years,thus the corrected value of the PV output is obtained.Afterward,by analyzing the difference between the corrected value and the actual value,the system dust accumulation state is evaluated,and a state alarm mechanism is also established.Finally,the effectiveness of this method is verified by experiments.(3)At present,the distributed PV O&M cycle is still mainly determined by manual experience,and the cost is too high,which is not conducive to the development of the distributed PV industry.To this end,this thesis proposes a method for determining the optimal O&M cycle considering typical scenarios of distributed PV systems.First,three mathematical models of distributed PV typical scenarios are established.The reliability model of PV components and the time-varying model of dust accumulation are also established.On this basis,the optimal O&M cycle determination model is established with the O&M cycles as the candidate solution set,and the combination of power generation loss cost and O&M cost as the objective function.Then,the Monte Carlo method is used to calculate the probability distribution of the total cost under different cycles,and the optimal cycle is obtained by calculating the average value of the simulation results.Finally,the effectiveness of the method in different seasons,grid-connected electricity prices,cleaning parameters,and PV operation years is verified through analysis of the examples.(4)Distributed PVs are widely distributed,located in complex terrain,and limited in O&M resources,which seriously increases O&M pressure.At present,there is still lack of effective large-scale PV O&M scheduling methods,resulting in untimely O&M and high costs.Therefore,this thesis first analyzes O&M tasks,personnel types,and equipment types.Second,a large-scale distributed PV O&M dynamic scheduling model is proposed.The scheduling model considers both planned tasks and random tasks.The model takes labor cost,equipment depreciation cost,power generation loss,maintenance downtime loss,and transportation cost as the comprehensive optimization objective.And it takes personnel skills,equipment utility,time window,etc.as constraints.Afterward,aiming at the problem that the traditional optimization algorithms are prone to fall into local optimum when solving large-scale optimization problem,a hierarchical hybrid of Genetic Algorithm and Discrete Binary Particle Swarm Optimization(HGA-BPSO)is proposed by combining the advantages of GA and BPSO.Finally,the effectiveness of the proposed model is verified by the case study. |
PDF Full Text Request