Research On Real-time Operation Model Of Three Gorges Reservoir Based On Deep Learning

Water is a clean energy source,and hydropower is developing rapidly in China.In recent years,the world-famous Three Gorges,Xiluodu and other large hydropower stations have been built.Reservoir scheduling is an important part of reservoir operation and management.With the large-scale hydropower development,reservoir scheduling needs to integrate hydrometeorology,power generation and other factors.It is difficult for traditional methods to consider multiple constraints and comprehensive goals in a short time.In previous studies,the design and extraction of scheduling functions and scheduling rules are mostly one-dimensional data information.The neural network structure is designed to be single,and it is difficult to deal with high-dimensional tensor data information.In view of the shortcomings of traditional scheduling decision-making methods,this paper proposes to use for reference in the field of image recognition Convolutional neural network(CNN)has been successfully applied,and the establishment of a convolutional neural network model suitable for reservoir dispatching has potential laws in hydropower dispatch itself.Therefore,this article takes the Three Gorges Power Station as the research object,and based on the current deep learning hotspot technology,deep learning Combining with reservoir flood control dispatching,comprehensively considering factors such as multiple constraints and application scenarios,simulating the Three Gorges Reservoir flood control dispatching,and generating training sample data.Establish a deep neural network,generate sample data based on simulated scheduling,perform network training,and build a real-time scheduling model for the Three Gorges Reservoir.The model can generate scheduling plans in real time and provide technical support for scheduling decision-makers.The main work content and innovative results of the thesis are as follows:(1)In view of the large area under the control of the Three Gorges Reservoir,the complex flood composition,the inconspicuous flood staging,the rapid flow change,and the uncertainty of the inflow,the reservoir operation simulation can fully consider the comprehensive utilization of the reservoir and the constraints,and the simulation is generated Reservoir scheduling process in different scenarios.Therefore,a real-time reservoir dispatching calculation model is established based on actual engineering data.The model generates a large number of reservoir simulation dispatching sample data,which provides a large number of training samples for the deep learning real-time dispatching model established later.(2)Neural network can find valuable information from various forms of data and then serve for decision-making.This paper takes the Three Gorges Reservoir as the research object,establishes a real-time dispatching model of the Three Gorges Reservoir based on deep learning,and extracts and learns the characteristics of real-time dispatching methods through deep neural network parameter training based on the training sample data generated by the simulated dispatch of the reservoir.In addition,a reinforcement learning algorithm is added to the deep learning model.The model optimizes model parameters through continuous optimization of reward strategies,and uses reinforcement learning training methods to achieve the final scheduling decision.The case study results show that the deep learning model has good convergence and can be applied to reservoir operation.Therefore,the model is used to make the real-time operation plan of the Three Gorges Reservoir.