Energy-Saving Oriented Optimization And Real-Time Regulation Of A Parallel Pumping System Based On The Modified Swarm Intelligence

Parallel pump systems have great advantages in terms of reliability and flow range,thus are widely applied in essential industries such as municipal sectors,chemical production,construction,electricity supply,and agriculture.The efficient and stable operations of such devices are of great significance for energy saving,reducing consumption,and safe production.However,compared to the advanced international technologies,the domestic efficiency-enhancing approaches still have about 20%of the improvement of space.To be specific,improving the efficiency and stability of parallel pump systems can mainly be achieved through hydraulic optimization of the pumps and optimization of the regulation strategies of the pump system.Nevertheless,based on the current computing power,the efficient solution to these two types of optimization problems is still limited by the model accuracy and algorithm performance.In order to change this situation,this dissertation conducts a study on the main problems in pump hydraulic optimizations and real-time regulation of parallel pump systems,funded by Na-tional Natural Science Foundation of China(Grant No.51879121),et al.Firstly,to address the drawbacks of the global searching algorithms with high computational costs,this dissertation pro-poses a modified particle swarm optimization algorithm based on particle classification and adap-tive acceleration strategy,which has remarkable advantages in convergence capability and search-ing accuracy.The modified algorithm is applied in the hydraulic optimization of an inline pump and the energy efficiency enhancement of the parallel pump system.Meanwhile,for the fact that there is no basis for the selection of the surrogate model in pump hydraulic optimizations,this dissertation provides a quantitative performance comparison of three surrogate models which are popular in recent years.Furthermore,a real-time regulation system based on artificial intelligence and swarm intelligence is proposed to improve the overall performance of the pump system in the face of varying system demands.The main contents and results of this dissertation are given as follows.(1)The progress of domestic and international research on pump hydraulic optimization and energy-saving oriented system regulation problems is systematically summarized,and the ad-vantages and applicabilities of different methods are compared and analyzed.Based on the current level of computational power,the method based on surrogate models or system models is still the one with the best comprehensive performance.(2)This dissertation innovatively proposes a modified particle swarm optimization algorithm based on particle classification and adaptive acceleration techniques,which is designed to have outstanding convergence capabilities.Meanwhile,a self-learning operator is added to improve the ability to escape from local optimums and enhance stability.Specifically,the population will be divided into three categories in accordance with fitness,namely leaders,elites,and explorers.Dif-ferent velocity and position update strategies are applied to these particles to balance the exploita-tion and exploration abilities and to maintain the population diversity during the search process.In order to verify the performance of the proposed algorithm,two sets of benchmark functions with varying complexity are adopted to evaluate the convergency capability,accuracy,and stability.Moreover,several state-of-the-art PSO variants are selected for comparison with the proposed al-gorithm in the benchmark tests.The results show that,compared to other participating algorithms,the proposed algorithm shows a significant advantage in search speed and can maintain excellent accuracy and stability even in the face of highly complex problems.(3)This dissertation conducts a quantitative performance comparison of response surface models,Kriging models,and artificial neural networks in multiparameter pump optimizations based on 3400 sets of high-precision samples,which aims to provide a reference for the selection of the surrogate model and sample size in subsequent studies.The results show that,for complex hydraulic optimizations,the response surface model cannot meet the requirements of engineering applications in all performance indicators.In contrast,the artificial neural network and the Kriging model perform better,while the Kriging model has higher requirements on both sample size and diversity.Specifically,the minimum sample size requirement for the artificial neural network is about 0.8 times the number of model parameters,and the figure for the Kriging model is about ten times the number of decision parameters.(4)This dissertation innovatively designs an induced vane before the impeller for an indus-trial inline pump,which is proven to suppress the inflow distortions that contribute to the deterio-ration of the hydraulic performance.Meanwhile,a parametric design method for such a structure is proposed based on the non-uniform rational basic spline,and multi-objective optimization is carried out using a modified PSO algorithm and artificial intelligence.In addition,in order to study the performance improvement of the novel proposed inlet structure,a comparative analysis is per-formed with an excellent model reported in the previous research using hydraulic loss visualization techniques.The results show that,within strict size constraints,the correlation between the shape of the inlet pipe and the performance of the inline pump is small and concentrated in the second half of the bend(near the outlet).At the same time,the positions of the leading-edge and trailing edge of the induced vane significantly impact the pump performance.Moreover,the CFD analysis results prove that the addition of the induced vane can effectively suppress the backflow vortices near the outlet of the inlet pipe.In contrast,its effectiveness in suppressing the development of flow separation vortices is not satisfactory.Better results can be achieved by refining the shape of the inlet pipe.Furthermore,enlarging the longitudinal size of the inlet pipe offers significant ad-vantages in reducing hydraulic losses in almost all aspects.However,when installation space is severely limited,similar results can be achieved by optimizing the inlet pipe shape and adding an induced vane.After optimization,the efficiency of the inline pump increased by 1.10%,1.59%,and 1.53%at 0.8Q_d,1.0Q_d,and 1.2Q_d,respectively.(5)In this dissertation,a pump test system in parallel configuration with fully automatic mon-itoring capabilities and comprehensive regulation abilities is designed and established.Parametric analysis and energy-saving oriented optimization are conducted based on the modified PSO algo-rithm and machine learning techniques.The results show that artificial neural networks show in-credibly high precision in predicting the characteristics of the pump system with a low sample demand and can effectively avoid the errors caused by the failure of the affinity laws under off-design operational conditions.Meanwhile,the parallel pump system is proven to have a vast en-ergy-saving potential of 70.56%at 50%system load.However,the energy-saving potentials achieved by system regulations will decrease gradually along with the increased system load.In addition,it is worth mentioning that the most efficient system status does not equal the lowest system energy consumption,and it always costs more power to consider the system reliability at off-design operational conditions.On the other hand,the optimization process also compares and analyzes four common regulation targets.In general,minimizing the system power consumption with the limitation of pump operational interval is most recommended as the objective function since the comparison of optimization results shows that this strategy performs well in reducing energy consumption,improving system stability and lifetime,and controlling excess flowrate.In contrast,Minimizing the system excess flowrate is not a good choice for either energy saving or system stability improvement.(6)Based on the machine learning method,improved swarm algorithm,and fuzzy PID con-troller,this dissertation innovatively proposes a real-time regulation system for parallel pumping systems,and conducted a comprehensive test using the test rig with four pump units in parallel configuration.The results show that the intelligent regulation strategy can further bring about 38%energy saving potentials compared with the traditional regulation strategy,and the system power distribution curve of the intelligent regulation results has better continuity and smoothness.In ad-dition,the data of the model predictions during the test showed that the stability and accuracy of the machine learning approach in fitting the pump system characteristics did not degrade signifi-cantly in the face of a continuously increasing sample size,and the model performance could reach the application standard when the sample size was higher than 20.