Intelligent Water Visualization And Early Warning System Based On Water Consumption Prediction

Urban water consumption is an indispensable part of our life.Accurate prediction of urban water consumption contribute to intelligent water early warning,which has great practical significance and economic benefits.Characteristics of water consumption data are usually ignored in most of existing water consumption prediction methods.Generally these methods are too simple to simulate more complex mathematical operations.In addition,most of the current intelligent water early warning system is mainly for the simply collecting,displaying data and setting a fixed water consumption threshold.Therefore,water points prediction and intervals prediction for alarm and early warning are combined with data visualization to assist water analysts to quickly locate outliers.Firstly,the visualization early warning system was designed according to the process of abnormal analysis by water analysts.A feasible and practical method was proposed to receive and preserve real-time data,then the abnormal data was cleaned and the original data was normalized.Secondly,based on the results of preprocessing,in order to meet the requirements of refinement and operational efficiency,a shape-based piecewise aggregate approximation(SPAA)method was proposed to reduce the dimension of high-dimensional water consumption curve.It is because traditional Euclidean distance-based clustering algorithm is prone to problem that ignoring the shape features of the curves that a k-shape clustering algorithm based on sequence morphological similarity adaptive clustering numbers was proposed.In addition,a centroid-based clustering center calculation method was adopted to extract the water consumption curve shape of each type of cluster.Then,based on the result of clustering,a sparse continuous deep belief network(LSC-DBN)was proposed for point prediction of water consumption.This model introduces a sparse regularization term to continuous restricted Boltzmann machine(CRBM)with Gaussian distribution,which solves the problem of homogenization features and is also suitable for the input of water consumption data.Then,a dynamic interval prediction algorithm was proposed to solve problem that the fixed alarm threshold of the existing alarm system cannot effectively early warning.and a dual warning and early warning mechanism was proposed based on the point prediction and the actual value.Finally,the intervals prediction and the points prediction value were used to assist the data anomaly detection,and reasonable and practical data visualization was carried out for water information.The experimental results have shown that the SPAA-k-shape algorithm proposed in this paper can effectively reduce the dimension and reduce the clustering calculation time,which is suitable for the water curve clustering.The LSC-DBN model can accurately predict the short-term urban water consumption,which is better than the existing prediction models.At the same time,the prediction of intervals is accurate,dual alarm mechanism and data visualization improve the efficiency of abnormality detection by water analysts,which reduce false negatives and false positives.The intelligent water system received a favorable rating on the line.