| Concentrated solar power is clean and reliable,with large energy storage capacity,and can realize 24-hour continuous and stable power generation,which is an important part of the Chinese power generation industry to achieve peak carbon dioxide emissions and carbon neutrality.The control of the thermal collecting field of the solar thermal power station directly affects the safety,reliability,and power generation efficiency of the photothermal power station.Therefore,it is of great significance to study the thermal collecting system of the solar thermal power station.In this thesis,aiming at the randomness,complexity,and strong disturbance of the heat collecting system,soft sensing modeling and sliding mode predictive control are adopted to study the outlet molten salt temperature.Finally,the experimental data of the Dunhuang Dacheng 50 MW molten salt linear Fresnel heat collection system are used for verification,and good results are obtained.Firstly,due to the complexity of the heat collection system,some key parameters cannot be measured in real-time.A global ensemble learning soft sensor modeling method based on the Gaussian process regression(GPR)is proposed.The main factors affecting the heat collection loop are analyzed,and the input-output relationship of the thermal collection system is obtained;Then,the data that have been connected to the grid are extracted from Dunhuang solar thermal power station,and the historical data are divided into different subsets and subspaces by bootstrap and partial least squares regression analysis.For each subspace,different local domains are constructed by Gaussian mixture model clustering,and the diversified base model is established;The base model is fused to obtain the output model,which is the global ensemble soft sensor model.The global ensemble soft sensor model is applied to the Linear Fresnel heat collection circuit of Dunhuang molten salt,and the predicted output is compared with the measured value.The results show that the global ensemble soft sensor model has higher prediction accuracy than the single GPR model.Secondly,aiming at the problems that some models with poor estimation performance reduce the accuracy of model prediction and increase the complexity of the model,a selective ensemble learning soft sensor modeling method based on GPR is proposed.Ensemble pruning based on a genetic algorithm is used to remove the base models with poor performance,so as to ensure the diversity and accuracy of the base model,and then the base models with good estimation performance are fused.Due to the randomness of solar radiation intensity,ambient temperature,and wind speed in the heat collection system,the performance of the obtained prediction model will be degraded.Therefore,the model adaptive control is introduced to obtain the selective ensemble learning soft sensor model.The predicted output of the model is compared with the measured value,and the results show that the performance of the selective integration is better than that of the fully integrated soft sensor model,the model with adaptive control has better estimation performance.Finally,aiming at the randomness and strong disturbance of linear Fresnel solar thermal power collection system,a sliding mode predictive control strategy is proposed.Considering solar radiation intensity,molten salt inlet temperature,and ambient temperature,a dynamic mathematical model of the heat collecting loop is established,and a sliding mode predictive controller for the molten salt outlet temperature in the heat collecting field is developed based on the model.The model predictive control,sliding mode control,and sliding mode predictive control are compared.The results show that compared with the model predictive control and sliding mode control,sliding mode predictive control has a faster tracking speed and smaller tracking error for the outlet temperature of the heat collecting loop,and improves the robustness and anti-interference of the heat collecting system. |
PDF Full Text Request