Research On Photovoltaic Power Generation Forecasting Model Based On Improved Echo State Network

With the global awareness of sustainable development and environmental protection,more and more countries are replacing traditional energy with clean energy.Solar energy is considered as one of the most ideal clean energy sources due to its wide distribution and renewability.Photovoltaic(PV)power generation is affected by various factors,and there are still difficulties in integrating PV power generation system into traditional power grids.Therefore,how to predict accurately photovoltaic power generation has become a hot research topic.To solve this problem,echo state network(ESN)is used for PV power generation prediction,which has the advantages of low-dimensional space mapping to high-dimensional space and simple training process.To further improve the prediction performance of the model,some different improvement for the single-input single-output model are given in this paperFirstly,due to the large number of internal neurons in the reservoir of the single-input singleoutput ESN,the model calculation complexity is high and it is easy to encounter the problem of ill-matrices,leading to inaccurate fitting of the target curve.Therefore,an improved echo state network with three reservoirs is proposed,which utilizes the decomposition mechanism of the reservoir to decompose a single reservoir into three sub-reservoirs of different sizes.Combining with the fluctuating characteristics of photovoltaic power generation systems,a dual-feedback mechanism is introduced to enable the model’s output to receive timely and effective feedback correction.Finally,the validity of the model is verified by selecting a set of numerical simulation and two different actual PV data sets in different regions.Secondly,considering the correlation between the PV power generation in a certain period and the previous power generation in multiple periods in the actual system,a fractional-integerorder echo state network(FIO-ESN)is proposed for PV power generation prediction.Due to the local characteristics of fractional-order equations and the strong robustness of integer-order equations,the FIO-ESN can fully reflect the characteristics of PV power generation systems.To ensure that the model can run stably in actual prediction,a sufficient stability criterion is given.Finally,the practicality of the model is verified by selecting a set of simulation experiments and three different actual PV data sets in different regions.Finally,considering that the current improvement of ESNs is single-input single-output structure,and considering the factors that actually affection power generation,a fractional-integerorder broad echo state network(FIO-BESN)is given.The fractional-order reservoirs and integerorder reservoirs of each module in the model are connected in series to achieve deep feature extraction for multiple input variables.This design improves the generalization ability and stability of the model,and reduces the model computation complexity.The effectiveness of the FIO-BESN is verified through actual PV data sets.