Research On Stochastic Optimization Scheduling Of Integrated Energy System Based On Multi-task Prediction

The integrated energy system can effectively promote the complementary and mutually beneficial development of multiple energy sources,as well as the coordinated optimization of source,grid,load and storage.It promotes energy efficiency improvement and renewable energy consumption while ensuring energy security,and is an important lever for achieving sustainable economic and social development and dual carbon goals.However,significant uncertainties in energy production,use,and trading pose challenges to the economic,safe,and reliable operation of the integrated energy system.Therefore,this thesis quantifies the uncertainty of the integrated energy system through source-load-price multi task joint prediction and scenario method,and further obtains the system scheduling plan through stochastic optimization method.The specific research content is as follows:In terms of multi task joint prediction,this thesis first considers three types of uncertainties:renewable energy generation,multiple loads,and energy prices,in response to the demand for multi task prediction and the completeness of prediction objects in the integrated energy system.Then,based on the results of cross correlation and autocorrelation analysis,a dual layer sourceload-price joint prediction model based on multi task learning framework and CNN-AM-LSTM network was studied.The results show that proposed model reduces the workload under multi task prediction requirements while ensuring prediction accuracy,and improves the generalization of the prediction model.In addition,further research has been conducted on the attention mechanism in the model.Based on the convolutional block attention module,an improved sequential convolution attention module with a mixed channel and sequential attention mechanism has been proposed,effectively improving the prediction accuracy of the model.In terms of scenario analysis method,this thesis takes wind power generation with strong randomness and volatility as the research object.Firstly,a scenario generation method combining wind power state transition and prediction error characteristics is studied to address the difficulty of balancing volatility and serial correlation in generating scenarios.This method first converts the prediction curve into a series of probabilistic states based on the state transition matrix and extracts the state sequence scenario.Then,based on the error prediction box and multivariate inverse transformation sampling method,scenarios are extracted from the set of state sequence scenarios.Then,based on the optimal scenario reduction model,an improved simulated annealing algorithm is proposed to obtain the set of optimal scenarios,which further emphasizes the processing of edge scenarios while selecting representative scenarios.The calculation results show that the generated scenario combines the volatility and correlation of wind power,and the scenario reduction algorithm has better reduction accuracy and stability.In terms of stochastic optimization method,this thesis conducts research on the coordination and response strategies of supply and storage equipment to uncertainty,as well as the flexibility indicators of scheduling plans.Under the background of electric and heat integrated energy system,a two-stage stochastic optimization day ahead scheduling method based on the suppression of random fluctuations is studied.In the first stage,the economic optimal dispatching plan is obtained based on the day ahead prediction.In the second stage,multiple random scenarios are considered and considered as fluctuations,and the random fluctuations of the scenarios are suppressed by calling the energy supply and storage equipment in the system.The results show that this method can coordinate the randomness of energy storage and supply equipment in response to wind power,electric and heat loads,and improve the flexibility of scheduling plans.