Research On Interval Forecasting Of Renewable Energy Time Series From The Perspective Of Multi-Objective Optimization

In order to achieve the strategic goals of "peak carbon emissions" and "carbon neutrality," the vigorous development of renewable energy has become a trend in constructing a new generation of low-carbon and clean electricity systems.In the comprehensive establishment of the new power system,accurate prediction,error analysis,and improvement of forecasting accuracy for a series of relevant time series data,such as renewable energy output and power,play a crucial role.Interval forecasting serves as an effective means to quantify prediction uncertainties and also acts as a prerequisite input for optimization decision models of power systems under uncertain conditions,receiving extensive attention in the academic community.However,the generation of renewable energy exhibits significant intermittency and randomness,while the chaotic characteristics of atmospheric systems make it challenging to accurately predict the output levels of new energy sources,posing severe challenges to the secure operation of the power system.On one hand,the probability statistical features of prediction errors are highly complex,and traditional uncertainty prediction methods commonly rely on parameterized prior assumptions,making it difficult to achieve a balance between prediction accuracy and the computational complexity of multidimensional inputs.On the other hand,the lack of reliable prediction information for grid operators in formulating forward-looking decisions results in the infeasibility and sub-optimality of system operation strategies,thereby facing uncertainty in the electricity supply-demand situation.Therefore,precise quantification of the prediction uncertainty of renewable energy generation and the construction of decision mechanisms with broad adaptability to prediction uncertainty are of profound significance in promoting the friendly integration and large-scale accommodation of renewable energy.These issues represent one of the key challenges currently faced by new energy power systems.Addressing the aforementioned issues,this study adopts a multi-objective parameter optimization perspective to develop an uncertainty ensemble-interval prediction system by integrating multiple prediction algorithms for various renewable energy time series.Building upon the framework of "Constructing and Applying Interval Prediction Systems in Wind Energy,Wave Energy,and Tidal Energy Fields," the study employs data preprocessing strategies to classify the seasonality of irregular data and eliminate highfrequency noise from the original data,thereby improving the predictability and reliability of the prediction module.Neural network deep learning models and time series statistical models are utilized to enhance adaptability to complex data,capturing and analyzing characteristics of both linear and nonlinear data to acquire comprehensive information.Core parameters and the prediction system are optimized through multi-objective optimization algorithms.The distribution and weight coefficients are optimized to obtain Pareto-optimal weights.The developed prediction system enriches the research theories and methodologies for uncertainty prediction of renewable energy time series.The research findings provide theoretical support for the sustainable development of new energy generation and offer potential technical references for grid dispatching.This article consists of six chapters.Chapter 1 provides an introduction to the research background,rationale,framework,and significance,concluding with a discussion on the main innovations and limitations of this study.Chapter 2 elaborates on the research foundations,systematically reviewing the current state of renewable energy time series prediction,including uncertainty prediction models,multi-objective optimization applications,and a retrospective analysis of deterministic point prediction and uncertainty interval prediction in wind energy,wave energy,and tidal energy fields.The existing research is organized,summarized,and synthesized,incorporating performance evaluation metrics and hypothesis testing methods to validate the predictive performance of models.Chapters 3 to 5 form the main body of this research.These chapters,from a multiobjective optimization perspective,employ different algorithmic strategies and conduct empirical studies to establish renewable energy interval prediction systems tailored to the characteristics of various renewable energy time series data.In Chapter 3,an uncertainty decomposition prediction ensemble system is constructed for wind speed time series data.It incorporates an optimal submodel adaptive selection strategy,variational mode decomposition technique,and a newly proposed multi-objective vulture search optimization algorithm.The effectiveness of the developed decomposition-predictionensemble system is demonstrated from multiple aspects,including objective submodel selection,optimization of distribution function parameters,and evaluation of model significance,stability,and computational complexity.Chapter 4 focuses on wave energy time series data and presents an uncertainty combination prediction system.It integrates a combination prediction module of multiple prediction algorithms,fuzzy information granulation method as a data preprocessing technique,and the emerging multi-objective grasshopper parameter optimization algorithm.Through empirical analysis,the chapter demonstrates the performance of the system in terms of information fuzzy classification,Pareto-optimal weights in multi-objective optimization,and capturing and analyzing both linear and nonlinear information for prediction accuracy and stability.In Chapter 5,an interval upper-lower bound combination prediction system is constructed for tidal energy time series data.It incorporates a combination prediction module consisting of six interval prediction models,the singular spectrum analysis method as a data decomposition strategy,and the multi-objective Levenberg-Marquardt algorithm optimized by an initial population.After data preprocessing,the improved multi-objective optimization algorithm combines the upper and lower limits of individual model predictions to obtain the optimal values of the final prediction interval bounds.The predictive performance of the developed uncertainty prediction system is demonstrated from various perspectives.Chapter 6 provides a summary of the main conclusions of the research and offers prospects for future research directions.The main research contributions of this article can be summarized as follows:First,addressing the irregular fluctuations in wind speed data and the limitations of single prediction algorithms in capturing wind speed uncertainty,a decomposed adaptive optimal model selection ensemble prediction system was established to simulate experiments using actual wind speed data from wind farms.Variational mode decomposition technique was introduced to decompose the original wind speed time series data,removing high-frequency noise and effectively improving the clarity and predictability of the data.The optimal submodel selection strategy enhanced the adaptability of the prediction module to complex data.The newly proposed multiobjective vulture search algorithm effectively optimized the weight coefficients.This system not only improved the accuracy of point prediction but also quantified and analyzed the uncertainty of wind speed,thereby enhancing prediction accuracy and reliability.Second,addressing the irregular fluctuations and complexity of significant wave height data,as well as the presence of abnormal waveforms and other special data points,traditional physical prediction models were unable to effectively utilize and analyze all the information.A combination effective wave height multi-step point-interval prediction system based on the fusion of multiple prediction algorithms was established to simulate experiments using actual half-hourly wave height data from different buoy locations.In the combination prediction module,the advantages of deep learning and neural network models were integrated,and the multi-objective grasshopper optimization algorithm was introduced to obtain Pareto-optimal weights.The fuzzy information granulation strategy was employed to classify each valid original data information point,eliminating noise and other interfering information,achieving optimal point and interval prediction accuracy,and accurately analyzing the uncertainty of point predictions.Experimental results demonstrated that the developed system significantly improved prediction accuracy and stability.Third,addressing the complex seasonal characteristics of tidal height data,where traditional physical point prediction models failed to identify all the feature information and the accuracy of interval prediction was limited by the results of point prediction,a tidal energy combination interval upper-lower bound prediction system was proposed.By utilizing data decomposition and reconstruction strategies,noise in the original data was eliminated.Multiple interval prediction algorithms were introduced to predict the upper and lower bounds separately.The proposed multi-objective weight optimization algorithm,optimized by an initial population,was employed to obtain the Pareto-optimal weights for the combination prediction module,thereby achieving optimal prediction accuracy.To validate the superiority of the proposed system,simulations were conducted using three tidal original datasets.The results demonstrated that the system quantified and analyzed the uncertainty of predictions and exhibited good prediction accuracy and stability.The main innovations of this article are as follows:First,in contrast to existing deterministic prediction models,this study introduces data preprocessing strategies,multi-objective parameter optimization algorithms,and a system prediction module that integrates multiple prediction models.This approach proposes a series of renewable energy time series uncertainty prediction systems from a multi-objective perspective,further enhancing the predictive performance of the systems.Second,a novel multi-objective optimization algorithm,the multi-objective vulture optimization algorithm,is proposed in this study.Additionally,an improved version of the multi-objective Levenberg-Marquardt optimization algorithm is introduced.Compared to widely used traditional multi-objective optimization algorithms,these proposed algorithms exhibit superior search and convergence speeds.They can obtain competitive optimization results,providing a new choice for addressing multi-objective optimization problems.Third,previous energy forecasting research has often focused on deterministic prediction while neglecting the importance of uncertainty prediction.Moreover,there has been a tendency to concentrate on subjectively selecting prediction models,which fails to overcome the limitations of individual prediction models.In this study,three uncertainty combination and ensemble prediction systems based on different strategies are proposed.An optimal submodel adaptive selection strategy is introduced,and multiple prediction algorithms are integrated into the prediction module based on data characteristics to enhance adaptability and improve prediction accuracy and stability for different renewable energy time series data.Fourth,previous prediction research has predominantly focused on the modeling aspect while overlooking the importance of data preprocessing prior to prediction.In this study,three data preprocessing techniques are introduced based on the characteristics of the original data.These techniques involve data decomposition,denoising,and integration.The proposed system exhibits more comprehensive and stable predictive performance compared to other models,thus further improving prediction accuracy and providing a theoretical basis for renewable energy prediction.Overall,this study introduces novel approaches to uncertainty prediction,multi-objective optimization algorithms,adaptive model selection,and data preprocessing,all of which contribute to the advancement of renewable energy forecasting methodologies.