Research On Hybrid Prediction Model Of Photovoltaic Power Based On Dendritic Neuron Network

With the gradual development of renewable energy,large-scale photovoltaic power generating has been implemented.However,due to the influence of weather circumstances,photovoltaic energy generation exhibits intermittent and fluctuating characteristics.Its extensive grid link complicates the operation and dispatching of the power system.In this context,reliable short-term photovoltaic power forecasting can significantly enhance PV usage and grid dispatch reliability.This thesis proposes a hybrid photovoltaic power prediction model based on dendritic neuron network using similar day selection,wavelet packet decomposition,and a biogeography-based optimization algorithm.The main research findings of this thesis are as follows:To begin,this thesis proposes a data processing strategy based on similar day selection(SDS)and wavelet packet decomposition(WPD)to address the rising difficulty of prediction caused by the randomness and fluctuation of photovoltaic output.On the one hand,the suggested SDS approach is utilized to filter the training set based on anticipated irradiance data in order to increase the training set’s similarity.On the other hand,the photovoltaic sequence is deconstructed using the WPD algorithm to achieve sequence stabilization.The simulation results demonstrate that the suggested strategy is successful at gathering highsimilarity training samples,reducing the volatility of PV sequences,and lowering the complexity of prediction.Second,to address the low prediction accuracy and slow convergence speed of standard artificial neural networks,this thesis proposes a photovoltaic power prediction method based on an upgraded dendritic neuron model(DNM).While retaining the distinctive dendritic structure of DNM,this technique substitutes a simpler linear function for the complex sigmoid function of the DNM synaptic layer,hence improving prediction accuracy and convergence speed concurrently.The simulation findings indicate that,when compared to other established methods,the suggested method achieves a greater level of prediction accuracy and a faster convergence rate.Then,in order to address the issues that the classic DNM learning algorithm is prone to overfit and is easily influenced by the initial value,an improved biogeography-based optimization(IBBO)algorithm is presented for optimizing DNM.Among these,this thesis significantly enhances the optimization performance of the IBBO algorithm by optimizing the migration model,migration operator,and mutation operation,allowing it to more effectively complete the training of DNM weight and threshold.The simulation findings indicate that,when compared to the traditional learning algorithm and the other six optimization algorithms,the IBBO suggested in this thesis is more capable of training DNM.Finally,the usefulness and superiority of the photovoltaic power hybrid prediction model based on SDS-WPD-IBBO-DNM suggested in this thesis are demonstrated by comparing and analyzing the prediction results with those of existing single and hybrid models under various weather situations.The simulation findings demonstrate that the suggested short-term photovoltaic power hybrid prediction model has a greater prediction accuracy and a faster convergence rate,and is capable of efficiently dealing with complicated and changing photovoltaic power prediction scenarios.