Long Term Runoff Forecast Research Of Fengman Reservoir Based On Data Fusion

China's water resources are unevenly distributed in time and space.Climate change and human activities lead to frequent droughts and floods,which have become the main factors restricting economic development.River runoff plays a leading role in the water cycle system,and extreme runoff will form catastrophe.Runoff prediction is of great significance and value for flood control and drought relief,water resource planning and management.There are many influencing factors and complex changing characteristics of river runoff.The key of runoff prediction is to excavate the law of factors influencing runoff formation based on genetic analysis.The prediction period of medium and long-term runoff is long,the prediction accuracy is low,the formation mechanism of runoff is not clear,the analysis of single scale factor and the improvement of single statistical prediction method can not further improve the accuracy of runoff prediction,and hydrological workers do not dare to report extreme values,and the results of medium and long-term runoff prediction can only be used as a reference for practical work.The research on the theory and technology of long-term runoff forecast,the fusion of multi-scale factors and multi-method forecast results,further improving the accuracy and level of forecast,can provide support for reservoir operation,water resources development and utilization,etc.This paper takes the annual runoff of Fengman reservoir basin as the research object,selects astronomical,global and basin scale factors,analyzes the similarity,teleconnection,commensurability,structural characteristics and other laws between the mining factors and the water from the basin,and studies and improves the technical methods such as intelligent learning method,fuzzy reasoning method,astronomical factor comparison method,point aggregation diagram method,commensurability method and commensurability network structure method.A long-term runoff forecast model is established,which includes multi-scale factor information fusion including factor fusion,result fusion and structure fusion.The research results can effectively improve the accuracy of runoff and extreme runoff forecast in Fengman reservoir basin,and provide technical support for Fengman reservoir operation.The specific research results are as follows:(1)Using the statistical analysis method,the response laws of three scale factors and the characteristics of the water coming from the basin are explored.The results show that there are good statistical laws between the wet and dry state of the water coming from Fengman reservoir basin and the cold and warm characteristics of ENSO events,the distance between the occurrence time of ENSO events and the flood season,and the meteorological factors selected based on the agricultural proverbs,all of which can pass the hypothesis test.Based on linear correlation coefficient method,mutual information theory method and correlation degree analysis method,this paper studies the correlation between astronomical factor,meteorological factor,astronomical factor+ocean atmosphere factor+meteorological factor and water from the basin.The results show that the correlation of meteorological factor is the strongest,that of ocean atmosphere factor is the weakest,and that of lunar declination angle and water from the basin is the largest.(2)Based on the combination of factors obtained from correlation analysis,neural network,support vector machine,decision tree,random forest and other intelligent learning methods are used to fuse factors to predict runoff.The results show that the regression prediction of water quantity is poor,and the three-level classification prediction is better;different prediction methods,the corresponding optimal factors and their combinations are different,and the elm and RBF neural networks are better in training and prediction performance and have strong robustness.The results of multi method optimal classification forecast are fused,and the accuracy of qualitative forecast is 89.5%.(3)The phase contrast method is used to integrate astronomical factors,marine atmospheric factors and their combination to forecast runoff.The results show that the accuracy of the quantitative prediction is 63.16%,and the qualitative prediction of 24 solar term lunar calendar date+sunspot relative number is the best,the accuracy is 63.16%.The phase contrast method has a strong ability to identify the wet and dry attributes of extreme coming year,but it is difficult to predict the normal year effectively.The accuracy of using the quantitative prediction results to deduce the water level is low.There are years that can't be distinguished by phase comparison method.The fuzzy inference method is used to further analyze the similarity of calculation factors based on the combination of factors obtained from correlation analysis,and the factors are fused to predict runoff.In order to improve the fuzzy reasoning method,TOPSIS fuzzy comprehensive evaluation method,similarity derivation method and factor in and out method are introduced.The results show that the robustness of fuzzy inference method based on similarity derivation is better than that based on turksen fuzzy inference method.The quantitative prediction of runoff is poor and the qualitative prediction of runoff is better.The results of their respective optimal qualitative prediction are fused,and the accuracy is 73.68%.(4)The primary and secondary factor comparison method is used to improve the single astronomical factor comparison method and the distributed fusion structure astronomical factor comparison method,and the fusion results are used to forecast runoff.The accuracy of qualitative prediction is 63.16%.Based on the analysis of the teleconnection between the ocean atmospheric factors,meteorological factors and the water coming from the basin,the forecast results are modified,and the astronomical factor comparison method is further improved,so that the forecast accuracy is increased to 73.68%.(5)Draw the point aggregation map of the three scale factors and the inflow of the basin,and integrate the results to forecast the runoff.The results show that the 24 solar term lunar calendar date and lunar declination angle point clustering map have good robustness,the relative number of sunspots is highly discrete and it is difficult to accurately divide their clustering intervals,and the accuracy of qualitative prediction of the three large-scale factor point clustering map is 63.16%,57.89%and 21.05%,respectively.The teleconnection law of ocean atmosphere factor,meteorological factor and incoming water is taken as the point aggregation map of this kind of factor,then the runoff forecast result is obtained,and the forecast result is integrated with the point aggregation map of astronomical factor,so that the forecast accuracy is increased to 73.68%.(6)Runoff is divided into general,extreme and extreme point structure,and extreme runoff is predicted by fusion structure.The results show that the prediction results of factor fusion and result fusion can be used as the general inflow structure to fuse the information of multi factor and multi method,and the accuracy of prediction is 84.21%;the improvement of point surface combination method and its up and down...