Low-cost direct drinking water pipeline leakage detection method based on constant-pressure water supply SMOTE-ENN-GWO-SVM model

Abstract

The invention provides a low-cost direct drinking water pipeline leakage detection method based on a constant-pressure water supply SMOTE-ENN-GWO-SVM model. According to the technical scheme, the method comprises the following steps that 1, operation data of the water supply frequency converter are collected, and the data are marked according to historical working conditions, namely leakage or normality; 2, due to the fact that the data set collected in the step 1 is unbalanced, an SMOTE-ENN algorithm is adopted to process unbalanced data; 3, using a GWO algorithm to iteratively optimize hyperparameters of the SVM, namely a penalty factor C and a kernel function parameter g; and 4, constructing the SVM model through the optimal hyper-parameter obtained in the step 3. And 5, performing pipeline detection by using the model in the step 4. Under the condition of constant-pressure water supply, when intelligent detection equipment is insufficient and detailed pipeline data cannot be obtained, the pipeline can be detected by using the system.

Description

Technical Field

[0001] The invention belongs to the field of direct drinking water pipeline leakage detection, and in particular relates to a low-cost pipeline leakage detection method based on a SMOTE-ENN-GWO-SVM model of constant pressure water supply. Background Art

[0002] Water supply pipeline leakage has always been a difficult problem faced by my country's water supply industry. It not only affects the normal water use of residents, but also wastes precious drinking water resources, especially the leakage of direct drinking water pipelines. Due to the high cost of direct drinking water, the leakage of its pipelines will cause great economic losses to water supply companies. Therefore, the early warning research on water supply pipeline leakage has always been highly valued.

[0003] In the early stage, the research on early warning of water supply pipeline leakage mostly relied on manual judgment based on real-time data collected by sensors. However, this method is subject to subjective factors and completely depends on the practical experience of technicians. In recent years, the development of intelligent algorithms has prompted domestic and foreign scholars to propose a variety of innovative prediction methods for accurate early warning of pipeline leakage to improve the efficiency and accuracy of pipeline maintenance and management. Most of these scholars use intelligent algorithms to measure the data collected by different types of sensor equipment installed in the pipeline system to evaluate and determine whether the pipeline is leaking.

[0004] Although these methods have achieved certain success in early warning of water supply pipeline leakage, the need to deploy a large number of different types of sensors and other detection equipment in the water supply pipeline system has resulted in high investment costs, and most of the equipment is affected by the level of intelligent technology, which has certain limitations, especially for direct drinking water systems with smaller pipe diameters. Unlike urban tap water, the direct drinking water system is a special form of urban public water supply system. It uses tap water or groundwater wells as water sources, establishes deep treatment water supply stations in certain water use areas, and arranges the water supply network as a full-circulation network to ensure excellent water quality and hygiene, and uses constant pressure water supply to supply water to water users in the surrounding area.

[0005] In summary, based on its constant pressure water supply characteristics, this method carried out a study on pipeline leakage early warning detection using intelligent algorithms based on the dynamic changes of equipment such as water supply inverters in direct drinking water supply pipelines. Summary of the invention

[0006] This invention proposes a low-cost leakage warning model and supervised learning algorithm based on the operating history data of the water supply inverter and water supply tank equipment of the regional direct drinking water supply station. First, the unbalanced data set of pipeline leakage is adjusted by the SMOTE-ENN algorithm, and then the two optimal hyperparameters of the support vector machine (SVM) are dynamically found by the GWO algorithm, and finally the SMOTE-ENN-GWO-SVM model is established.

[0007] A direct drinking water pipeline leakage monitoring method based on the SMOTE-ENN-GWO-SVM model, the method comprising the following steps:

[0008] (1) Dataset collection: Collect the operating status data of the existing water supply inverters at the direct drinking water supply station, and mark the pipeline data based on historical operating conditions, marking them as leakage or normal.

[0009] (2) Imbalanced processing of pipeline leakage data: The ratio of majority samples to minority samples in the sample set is adjusted. First, the minority samples are synthesized, and then some noise point data are eliminated.

[0010] (3) GWO parameter optimization: The GWO algorithm is used to iteratively optimize the hyperparameters of the SVM based on the fitness function.

[0011] (4) SVM model prediction: Based on the GWO algorithm, a set of optimal hyperparameters are obtained to construct an SVM to predict the pipeline data set.

[0012] The collection of the data set in the step (1) requires the deployment of a large number of different types of sensors and other detection equipment in the water supply pipeline system, resulting in a high investment cost. In addition, most of the equipment is affected by the level of intelligent technology and has certain limitations. Based on this, when the intelligent detection equipment is insufficient, since most of the direct drinking water adopts constant pressure water supply, the existing water supply frequency converter of the direct drinking water station can be used for detection. A pipeline leakage early warning detection based on the dynamic changes of the water supply frequency converter and other equipment in the direct drinking water supply pipeline is proposed.

[0013] In the step (2), the imbalance of pipeline leakage data is processed by 1) first using SMOTE to generate a minority of pipeline leakage samples, thereby increasing the diversity of minority class samples, and 2) ENN eliminates potential noise and boundary fuzzy samples, thereby improving the overall quality of the data set.

[0014] The specific steps of step (3) GWO parameter optimization are as follows: 1) dividing the training set and the test set; 2) data preprocessing; 3) parameter setting; 4) iteratively training the SVM and obtaining its fitness; 5) obtaining the optimal hyperparameters.

[0015] The specific steps of the SVM model prediction in step (4) are as follows: 1) constructing an SVM based on the optimal hyperparameters obtained by GWO optimization; 2) training the optimal SVM model; 3) detecting the optimal SVM model.

[0016] Beneficial effects of the present invention:

[0017] (1) In the case of constant pressure water supply, when there is no sensor or other detection equipment deployed in the pipeline system, based on the principle of constant pressure water supply in direct drinking water pipelines, water supply inverters and other equipment can be used to warn of pipeline leakage, which is a low-cost pipeline leakage detection method.

[0018] (2) The SMOTE-ENN-GWO-SVM model significantly improves the classification performance of SVM on imbalanced data sets by combining the data balancing technology of SMOTE-ENN and the hyperparameter optimization of the GWO algorithm, and has better performance in the final model performance (such as accuracy, precision, recall, etc.). BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Schematic diagram of constant pressure water supply

[0020] Figure 2 Flowchart of the SMOTE-ENN-GWO-SVM model

[0021] Figure 3 The proportion diagram of the dataset after adjustment using different imbalanced dataset algorithms

[0022] Figure 4 Comparison chart of SMOTE-ENN-GWO-SVM and other models’ accuracy

[0023] Figure 5 Comparison of F1 scores between SMOTE-ENN-GWO-SVM and other models

[0024] Figure 6 Figure 2 is the ablation experiment result of SMOTE-ENN-GWO-SVM model DETAILED DESCRIPTION

[0025] like Figure 1 As shown in the figure, direct drinking water supply adopts constant pressure water supply. When the pipeline leaks, the pressure transmitter will detect the water pressure drop in the pipeline and transmit this information to the control system. The inverter will try to adjust the speed of the water pump according to the feedback signal of the control system to maintain the water pressure of the system. Based on this, a low-cost direct drinking water pipeline leakage detection model is proposed, as shown in Figure 2 As shown, the specific steps are as follows:

[0026] (1) Use the SMOTE-ENN algorithm to adjust the imbalanced data set, for each minority sample x i , and obtain x by using the K nearest neighbor algorithm and the Euclidean distance formula as the standard i The five nearest minority sample points, for each randomly selected sample point x j , through the formula x new =x+rand(0,1)×(xi -x j ) to get a new sample x new , and use the Enn algorithm to remove outliers.

[0027] (2) The adjusted dataset is divided into training set and test set in a ratio of 7:3.

[0028] (3) Initialize a gray wolf population of size N and randomly generate the position of each wolf, that is, N sets of C and g hyperparameters.

[0029] (4) With the above N sets of hyperparameters, N SVM models were trained on the training set. At the same time, the fitness of each wolf was calculated according to the fitness function. The accuracy of the SVM model was evaluated using 10-fold cross validation. Ten scores were obtained and the average of the scores was taken as its fitness.

[0030] (5) Compare the fitness of each wolf in the population and select the three with the highest fitness as the leader wolves, namely α wolf, β wolf and δ wolf. (6) Based on the current positions of the three leader wolves, calculate the positions that the remaining gray wolves need to adjust due to the influence of α wolf, β wolf and δ wolf, and start moving.

[0031] (7) Determine whether the number of iterations has reached a predetermined upper limit. If not, continue to repeat step (4).

[0032] (8) The position (C, g) of wolf α after the Tth iteration is taken as the optimal parameters C and g, and the corresponding SVM classification prediction model is the optimal model. (9) The test set is tested using the above SVM model, and the test results are output.

Claims

1. A direct drinking water pipeline leakage monitoring method based on the SMOTE-ENN-GWO-SVM model of constant pressure water supply, characterized in that A low-cost detection method based on the principle of constant pressure water supply, the method comprises the following steps: (1) Dataset collection: Collect the operating status data of the existing water supply inverters at the direct drinking water supply station, and mark the pipeline data based on historical operating conditions, marking them as leakage or normal. (2) Imbalanced processing of pipeline leakage data: The ratio of majority samples to minority samples in the sample set is adjusted. First, the minority samples are synthesized, and then some noise point data are eliminated. (3) GWO parameter optimization: The GWO algorithm is used to iteratively optimize the hyperparameters of the SVM based on the fitness function. (4) SVM model prediction: A set of optimal hyperparameters of SVM obtained based on the GWO algorithm is used to train and predict the pipeline dataset.

2. A water supply prediction method based on the SMOTE-ENN-GWO-SVM model as claimed in claim 1, characterized in that: The collection of the data set in step (1) requires the deployment of a large number of different types of sensors and other detection equipment in the water supply pipeline system, resulting in a high investment cost. In addition, most of the equipment is affected by the level of intelligent technology and has certain limitations. Based on this, when intelligent detection equipment is insufficient, since most direct drinking water adopts constant pressure water supply, the existing water supply frequency converter of the direct drinking water station can be used for detection. A pipeline leakage early warning detection based on the dynamic changes of water supply frequency converters and other equipment in the direct drinking water supply pipeline is proposed.

3. The water supply prediction method based on the SMOTE-ENN-GWO-SVM model according to claim 1, characterized in that: In step (2), the imbalanced processing of pipeline leakage data is first performed by using SMOTE to generate a minority of pipeline leakage samples, thereby increasing the diversity of minority class samples, and then using ENN to eliminate potential noise and boundary fuzzy samples, thereby improving the overall quality of the data set.

4. The water supply prediction method based on the SMOTE-ENN-GWO-SVM model according to claim 1, characterized in that: Step (3) The specific steps of GWO parameter optimization are as follows: 1) divide the training set and the test set; 2) data preprocessing; 3) parameter setting; 4) iteratively train the SVM and obtain its fitness; 5) obtain the optimal hyperparameters.

5. The water supply prediction method based on the SMOTE-ENN-GWO-SVM model according to claim 1, characterized in that: Step (4) SVM model prediction The specific steps are as follows: 1) Constructing SVM based on the optimal hyperparameters obtained by GWO optimization; 2) Training the optimal SVM model; 3) Detecting the optimal SVM model.

Images