Research On Key Technologie Of Intelligent Operation And Maintenance Of Photovoltaic Power Station

In order to respond to the national strategic needs and help achieve the goal of"carbon neutral",China is changing from a traditional energy system to a new power system led by new energy generation,of which,photovoltaic as an important form of energy in the new power system,with the advantages of low pollution,wide sources and large storage capacity,the future of photovoltaic may become the main global energy supply situation.However,with the gradual expansion of the scale of photovoltaic power plants,there are problems of missing data and low prediction accuracy in the operation of photovoltaic power plants,therefore,intelligent operation and maintenance of photovoltaic power plants emerged as the times require.In order to improve the operational efficiency of PV power plants,promote the construction of new power systems,and help achieve the goal of "carbon neutral and carbon peaking",the research of hybrid storage capacity optimization and short-term power prediction method of PV power plants is carried out.Firstly,an intelligent optimal allocation method for hybrid storage capacity of PV power plants considering grid power leveling and economy is proposed.The hybrid storage capacity optimization model with weights is established by minimizing the economic cost of battery-flywheel hybrid storage and the optimal leveling index,setting constraints on the hybrid storage capacity optimization model,designing the hybrid storage charging and discharging control strategy,and using the political optimization algorithm(LSCAPO)that integrates the logistic chaos mapping and the sine cosine algorithm to find the optimal hybrid storage capacity optimization model.The feasibility of the proposed method is verified based on the real active power data of a photovoltaic power plant to provide ideas for hybrid energy storage applications.Second,the BOHB-Elman PV plant short-term power prediction method based on similar days is proposed.Based on the weather forecast and power data of PV power plants after data preprocessing,the power feature vector and meteorological factor feature vector are constructed,Gaussian mixture clustering is performed on the power feature vector,the similar day of the prediction day is selected using improved gray correlation analysis,and then the optimal values of the Elman neural network hyperparameters are determined using Bayesian and overband optimization algorithm(BOHB)to establish the PV power plant short-term power prediction model.The BOHB-Elman model improves the prediction accuracy by 3.83%compared with the existing model by relying on the actual power data and weather forecast data of a PV plant for short-term power prediction.Finally,a multivariate predictive graph neural network(DRGC-Iverted-MTGNN)incorporating adaptive nodal scaling parameters and inverse residual structure is proposed for short-term power prediction of PV power plants in order to make short-term power prediction with real-time and reliability.In DRGC-Iverted-MTGNN,the graph learning layer and the spatio-temporal convolution module are improved,the inverse residual structure is introduced and the node scaling parameters are adaptively adjusted to learn data features from the temporal and spatial levels.Based on the lab-built PV plant power intelligent prediction workstation,the model is compared and tested with various MTGNN improvement models,and the results show that the prediction accuracy of DRGC-Inverted-MTGNN is improved by 3.56%compared with the existing models.In addition,a data-driven short-term prediction software for active power of PV power plants was developed based on the Python language environment and tested in the laboratory environment to provide a solid foundation for practical engineering applications.