Water supply network leakage identification method based on particle swarm optimization BP neural network

By optimizing the BP neural network in combination with the pressure difference and average pressure characteristics of the DMA partition, the problems of overfitting and low efficiency in the leakage identification of water supply pipelines are solved, and efficient and low-cost leakage identification and positioning are achieved. In the extensive monitoring of water supply pipelines.

CN119084840APending Publication Date: 2024-12-06LIAONING TECHNICAL UNIVERSITY
View PDF 12 Cites 0 Cited by

Patent Information

Application Number
CN202411297316.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-18
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The prior art has problems of overfitting, computing speed and inefficiency in the identification of leakage loss of water supply pipelines, making it difficult to effectively identify and accurately locate the leakage position, and the manpower and material resources are invested, and the applicability is limited.

Method used

Using the method of optimizing BP neural network based on particle swarm, the prediction model is constructed by obtaining the flow data and pressure data of the water supply network, and combining the PSO optimization algorithm and the BP neural network to optimize the model parameters to achieve leakage identification and positioning, which is suitable for Pressure difference and average pressure characteristics of DMA partition.

Benefits of technology

Improves the accuracy and efficiency of leakage identification, reduces costs, and is suitable for leakage monitoring of water supply networks in general, avoids the local optimization and overfitting problems of BP neural networks, and improves the computer and deep learning. Application effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119084840A_ABST
    Figure CN119084840A_ABST
Patent Text Reader

Abstract

The invention discloses a water supply network leakage identification method based on a particle swarm optimization BP neural network. The method comprises the following steps: acquiring flow data and pressure data of a water supply network; performing partition division on the water supply network, and obtaining pressure characteristics according to a partition division result and the pressure data; a prediction model is constructed, the prediction model comprises an optimization algorithm model and a neural network model, the input of the deep learning model is feature variables, the feature variables comprise node flow data, pressure data and pressure features, and the output of the deep learning model is an output flow prediction value; based on the flow data, the pressure data and the pressure features, parameter adjustment and optimization are carried out on the deep learning model, and a final prediction model is obtained; and identifying the data of the water supply network to be detected through the final prediction model to obtain a flow prediction value of the water supply network to be detected, and judging the actually measured output flow value and the flow prediction value of the water supply network to be detected to obtain a leakage identification result of the water supply network.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of leakage area positioning, and in particular relates to a water supply network leakage identification method based on particle swarm optimization BP neural network. Background Art

[0002] Affected by the aging of the pipeline network, pipe quality, instrument accuracy, construction quality, stress, environmental corrosion, complex topological structure, etc., leakage in the water supply network is a common phenomenon, so it is urgent to reduce the leakage rate of the pipeline network.

[0003] Water supply network leakage not only causes a large amount of high-quality water resources to be wasted, but also affects the normal water use of industrial production and residents' lives. At the same time, the leakage of the network also affects the economic benefits of water supply companies to a certain extent. Therefore, for water supply companies, preventing and controlling water losses caused by network leakage is an important part of network operation management. Network leakage control is an important guarantee for achieving water conservation and consumption reduction and improving water resource utilization efficiency.

[0004] Common leakage control technologies include leakage identification and positioning based on leak detection instruments (audio-leak detection, ground-penetrating radar pipeline endoscopy, etc.), asset management, pipeline network optimization design, pressure management, etc. Acoustic leak detection technology uses correlation functions and acoustic sensors installed on pipelines or fire hydrants to measure the propagation time of sound waves. If the pipeline leakage point is located between two sensors, the sensor will detect the noise emitted by the leak, and then the time difference caused by the same leakage noise propagating to the two sensors can be determined based on the calculation results of the correlation function: when the sound speed in the pipeline is determined, the specific location of the leakage point can be calculated accordingly. However, this technology has many shortcomings: (1) It is not applicable to non-metallic pipelines; (2) The sensor layout density is high and the investment is high; (3) It is easily affected by environmental noise and nearby buried cables.

[0005] In recent years, with the development of sensor technology and computer technology, smart water services have gradually emerged. However, a large amount of working condition information data of the pipeline network has not been fully utilized and analyzed. At present, the water network monitoring technology based on traditional sensor data statistical analysis is difficult to effectively identify the occurrence of leakage, let alone accurately locate the location of the leakage. Leakage identification and positioning technology based on leak detection instruments is widely used in practice, which can effectively find leakage points and reduce the leakage rate of the pipeline network. This method is more accurate and intuitive, but it also requires more manpower and material resources. When applied on a large scale, the efficiency is low.

[0006] With the application of big data technology in the field of leakage identification, in-depth and systematic research on leakage problems, and active exploration of effective leakage control methods through systematic integration, information control and intelligent management of water data are of great significance to improving the economic and social benefits of water supply companies. Summary of the invention

[0007] In order to solve the above technical problems, the present invention proposes a water supply network leakage identification method based on particle swarm optimization BP neural network to solve the defects of the above-mentioned prior art that some big data algorithms are prone to overfitting, low calculation speed and efficiency.

[0008] To achieve the above object, the present invention provides a water supply network leakage identification method based on particle swarm optimization BP neural network, comprising:

[0009] Obtaining flow data and pressure data of a water supply network; dividing the water supply network into zones, and obtaining pressure characteristics based on the zone division results and the pressure data;

[0010] Constructing a prediction model, wherein the prediction model includes an optimization algorithm model and a neural network model, the input of the deep learning model is a feature variable, wherein the feature variable includes node flow data, pressure data and pressure characteristics, and the output is an output flow prediction value;

[0011] Based on flow data, pressure data and pressure characteristics, the deep learning model is adjusted and optimized to obtain the final prediction model;

[0012] The water supply network data to be tested and the measured output flow value are obtained, and the water supply network data to be tested are identified through the final prediction model to obtain the flow prediction value of the water supply network to be tested, and the measured output flow value and the flow prediction value of the water supply network to be tested are judged to obtain the water supply network leakage identification result.

[0013] Optionally, the water supply network is zoned by DMA zoning, and the pressure characteristics include regional pressure difference and average pressure, wherein the regional pressure difference is the difference between the mean value of the inlet pressure gauge data and the mean value of the outlet pressure gauge data in each divided area in the zoning result; the average pressure is the mean value of all pressure gauge data in each divided area.

[0014] Optionally, the neural network model adopts a BP neural network model, wherein the BP neural network model includes an input layer, a plurality of hidden layers and an output layer connected in sequence, wherein the hidden layer includes a plurality of hidden nodes.

[0015] Optionally, the optimization algorithm model adopts a PSO optimization algorithm.

[0016] Optionally, the process of building a prediction model includes:

[0017] Constructing a neural network model, and initializing the network structure and network parameters of the neural network model;

[0018] An optimization algorithm model is constructed, and individual dimensions in a population of the optimization algorithm model are matched with network parameters of a neural network model, wherein the individual dimensions in the population are equal to the sum of the number of weights and thresholds in the neural network model.

[0019] Initializing model parameters in the optimization algorithm model;

[0020] The objective function and fitness function of the optimization algorithm model are set, wherein the objective function is the error between the output of the neural network model and the actual value, and the fitness function is the mean square error of the output of the neural network model.

[0021] Optionally, the optimization process of the prediction model includes:

[0022] Integrate and label the flow data, pressure data, and pressure features to obtain training sets and test sets;

[0023] Iteratively input the training set, and in each iteration process, calculate the fitness of individuals in the population in the optimization algorithm model; according to the fitness of individuals in the population, update the population, and bring the updated weights and thresholds into the neural network model for parameter update, and perform the next iteration process based on the neural network model with updated parameters until the maximum number of iterations is reached, and the optimal solution of the population individuals is obtained, and the optimal solution of the population individuals is used as the initial value of the weight and threshold in the neural network model to obtain a pre-trained model;

[0024] The neural network model is trained and tested through a training set and a test set to obtain an optimized prediction model.

[0025] Optionally, the process of updating the population includes:

[0026] According to the fitness of individuals in the population, the current individual extreme value and the current group extreme value are obtained, and the population is updated according to the current individual extreme value and the current group extreme value.

[0027] Optionally, the process of tuning the prediction model includes:

[0028] The influencing parameters of the optimization algorithm model and the topological structure of the neural network model are iteratively adjusted. In each iterative adjustment process, the prediction model is optimized, and the model accuracy of the optimized prediction model is judged. According to the model accuracy judgment result, the hyperparameters of the prediction model obtained in each iterative adjustment process are adjusted until the judgment condition for ending the iteration is met to obtain the final prediction model.

[0029] Optionally, the accuracy judgment process includes: judging the model accuracy of the optimized prediction model through root mean square error and determination coefficient.

[0030] Optionally, the process of judging the measured flow value and the predicted flow value of the water supply network to be measured includes:

[0031] The difference calculation is performed on the measured flow value and the predicted flow value of the water supply network to be measured. If the difference calculation result is within the range of the model accuracy index and the normal leakage, it is considered that there is no abnormal leakage. Otherwise, there is an abnormal situation. The leakage point is detected and located according to the area where the flow meter corresponding to the measured flow value where the abnormal situation exists is located.

[0032] Compared with the prior art, the present invention has the following advantages and technical effects:

[0033] 1. The PSO-BP model for leakage prediction based on the identification of pressure difference and average pressure of DMA partitions in the present invention takes into account the main factors affecting pipeline network leakage, including regional pressure difference and regional average pressure, to ensure that the designed model is closer to the actual pipeline network leakage situation.

[0034] 2. The present invention mainly utilizes pressure data to realize leakage differentiation and identification, which is applicable to the situation that my country's water supply network generally only installs flow meters at regional entrances and exits and arranges a large number of pressure monitoring points, and has the characteristics of low cost and high practicality. By optimizing the initial weights and thresholds of the BP neural network model through PSO, it is possible to avoid the defects of the BP neural network easily falling into local optimality, overfitting, low computing speed and efficiency from the beginning, and can solve the defects of overfitting, low computing speed and efficiency that are prone to occur through some big data algorithms in the above-mentioned prior art. In addition, the present invention can be well matched with various prediction models. With the development of computers and deep learning, more excellent pipeline prediction models will further improve the final leakage identification results of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0036] Figure 1 A schematic diagram of the steps of a method for identifying leakage in a water supply network according to an embodiment of the present invention;

[0037] Figure 2 The present invention is a flow chart of a method for identifying leakage in a water supply network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0039] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0040] The present invention provides a method for identifying pipe network leakage based on DMA partition pressure difference and average pressure, thereby overcoming the defects of some big data algorithms in the prior art that are prone to overfitting, low calculation speed and efficiency, and providing a method for identifying pipe network leakage.

[0041] To achieve the above object, the present invention provides a water supply network leakage identification method based on particle swarm optimization BP neural network, comprising:

[0042] Obtaining flow data and pressure data of a water supply network; dividing the water supply network into zones, and obtaining pressure characteristics based on the zone division results and the pressure data;

[0043] Constructing a prediction model, wherein the prediction model includes an optimization algorithm model and a neural network model, the input of the deep learning model is a feature variable, wherein the feature variable includes node flow data, pressure data and pressure characteristics, and the output is an output flow prediction value;

[0044] Based on flow data, pressure data and pressure characteristics, the deep learning model is optimized and tuned to obtain the final prediction model;

[0045] The water supply network data to be tested and the measured output flow value are obtained, and the water supply network data to be tested are identified through the final prediction model to obtain the flow prediction value of the water supply network to be tested, and the measured output flow value and the flow prediction value of the water supply network to be tested are judged to obtain the water supply network leakage identification result.

[0046] As some embodiments, the water supply network is zoned by DMA zoning, and the pressure characteristics include regional pressure difference and average pressure, wherein the regional pressure difference is the difference between the average value of inlet pressure gauge data and the average value of outlet pressure gauge data in each divided area in the zoning result; the average pressure is the average value of all pressure gauge data in each divided area.

[0047] As some embodiments, the neural network model adopts a BP neural network model, wherein the BP neural network model includes an input layer, a plurality of hidden layers and an output layer connected in sequence, wherein the hidden layer includes a plurality of hidden nodes.

[0048] As some embodiments, the optimization algorithm model adopts a PSO optimization algorithm.

[0049] As some embodiments, the process of building a prediction model includes:

[0050] Constructing a neural network model, and initializing the network structure and network parameters of the neural network model;

[0051] An optimization algorithm model is constructed, and individual dimensions in a population of the optimization algorithm model are matched with network parameters of a neural network model, wherein the individual dimensions in the population are equal to the sum of the number of weights and thresholds in the neural network model.

[0052] Initializing model parameters in the optimization algorithm model;

[0053] The objective function and fitness function of the optimization algorithm model are set, wherein the objective function is the error between the output of the neural network model and the actual value, and the fitness function is the mean square error of the output of the neural network model.

[0054] As some embodiments, the optimization process of the prediction model includes:

[0055] Integrate and label the flow data, pressure data, and pressure features to obtain training sets and test sets;

[0056] Iteratively input the training set, and in each iteration process, calculate the fitness of individuals in the population in the optimization algorithm model; according to the fitness of individuals in the population, update the population, and bring the updated weights and thresholds into the neural network model for parameter update, and perform the next iteration process based on the neural network model with updated parameters until the maximum number of iterations is reached, and the optimal solution of the population individuals is obtained, and the optimal solution of the population individuals is used as the initial value of the weight and threshold in the neural network model to obtain a pre-trained model;

[0057] The neural network model is trained and tested through a training set and a test set to obtain an optimized prediction model.

[0058] As some embodiments, the process of updating the population includes:

[0059] According to the fitness of individuals in the population, the current individual extreme value and the current group extreme value are obtained, and the population is updated according to the current individual extreme value and the current group extreme value.

[0060] As some embodiments, the process of tuning the optimized prediction model includes:

[0061] The influencing parameters of the optimization algorithm model and the topological structure of the neural network model are iteratively adjusted. In each iterative adjustment process, the prediction model is optimized, and the model accuracy of the optimized prediction model is judged. The prediction model obtained in each iterative adjustment process is screened according to the model accuracy judgment result until the judgment condition for ending the iteration is met to obtain the final prediction model.

[0062] As some embodiments, the process of accuracy judgment includes: judging the model accuracy of the optimized prediction model through the root mean square error and the determination coefficient.

[0063] As some embodiments, the process of judging the measured flow value and the predicted flow value of the water supply network to be measured includes:

[0064] The difference calculation is performed on the measured flow value and the predicted flow value of the water supply network to be measured. If the difference calculation result is within the range of the model accuracy index and normal leakage, it is considered that there is no abnormal leakage, otherwise there is an abnormal situation.

[0065] In order to describe the above technical solution in detail, the above technical solution of the present invention is described in detail in combination with the accompanying drawings and related data:

[0066] like Figure 1-2 As shown, in order to achieve the above-mentioned purpose of the invention, the technical solution adopted by the present invention is:

[0067] A water supply network leakage identification method based on particle swarm optimization BP neural network includes the following steps:

[0068] S1. Obtain the flow and pressure data of the water supply network according to the production and operation status of enterprises in the study area;

[0069] S2. According to the study area, the water supply network is divided into zones by the DMA zoning method. Based on the zoning results, the pressure characteristics of the DMA zones are calculated, where the pressure characteristics include the regional pressure difference and the average pressure difference;

[0070] S3, using the four characteristic quantities of the DMA partitioned water supply network node flow, node pressure, regional pressure difference, and average pressure as input data, using the real output flow data as label data, randomly selecting 80% of the measured data set as the training set, and the remaining 20% ​​of the data as the test set;

[0071] S4, constructing a prediction model, wherein the prediction model adopts a PSO-BP model, inputting the divided training set and test set into the PSO-BP model, optimizing and tuning the constructed PSO-BP model until the model parameters are optimal, substituting the optimal model parameters into the BP model, and obtaining the output flow prediction value through the final prediction model;

[0072] S5. Identify abnormal conditions in the pipe network based on the output flow prediction value. Compare the output flow prediction value with the measured flow value of the flow meter. If the error between the predicted value and the measured value is within the model accuracy (R 2 ≥0.9), it is considered that there are no abnormal conditions such as pipe burst and serious leakage in the area measured by the flow meter; otherwise, it is considered that there are abnormal conditions in the area measured by the flow meter, and the leakage of the water supply network in the area is identified. The abnormal data is input into the model, and the accuracy R is used as the evaluation standard of the pipe network anomaly detection model.

[0073] As some embodiments, the regional pressure difference of the DMA partition in S2 is the mean of all inlet pressure gauge data of each sub-partition minus the mean of outlet pressure gauge data; the regional average pressure of the DMA partition is the mean of all pressure gauge data within each sub-region.

[0074] As some embodiments, the PSO-BP model construction method in S4 is:

[0075] Design the structure of the BP neural network model. The BP neural network model includes an input layer, several hidden layers and an output layer connected in sequence, wherein the hidden layer contains several hidden nodes. The network structure and network parameters of the neural network model that need to be set include: the input layer and the output layer, the initial value interval of the weight and threshold, the number of hidden layers, the number of nodes, the activation function, the learning rate, the maximum number of training times, and the system error;

[0076] Construct a PSO optimization algorithm model and establish the corresponding relationship between the particle swarm, weights and thresholds in the PSO optimization algorithm model. The particle latitude D in the particle swarm algorithm is the sum of the number of weights and thresholds in the BP neural network structure;

[0077] Initialize the model parameters of the PSO optimization algorithm model. Initialize and define the maximum number of iterations, population size, number of particles, particle speed and particle position, and set the model parameters of the PSO optimization algorithm model such as inertia weight, individual learning factor, and social learning factor;

[0078] Setting the objective function and fitness function of the optimization algorithm model, wherein the objective function is the error between the output of the neural network model and the actual value, and the fitness function is the mean square error of the output structure of the neural network model; looping the PSO optimization process, continuously updating the position of the optimal particle until the maximum number of iterations is reached, and terminating the PSO algorithm;

[0079] Set the population update method, determine the individual extreme value and the population extreme value, and update the particle speed and position. When the fitness value of an individual after updating its position is better than the current individual extreme value, it will be replaced by the new individual extreme value and the speed and position will be updated; when all individuals in the population update their speed and position, and any individual fitness value is better than the population extreme value, the minimum fitness value will be replaced as the population extreme value;

[0080] Set the optimal solution for the output particles. The optimal solution for the particles here is the best position of the particles, that is, the initial value of the weights and thresholds of the BP neural network after optimization.

[0081] Set the post-training mode of the BP neural network. After optimizing the BP neural network model through the PSO optimization algorithm model, use the training set and test set to train and test the BP neural network separately, and continuously adjust the weights and thresholds until the prediction error is within the set error range or the maximum number of training times is reached.

[0082] The construction process involves the model's initial structure, parameters, and operations. Through the above content, the model can be effectively optimized and the operation efficiency can be improved.

[0083] As some embodiments, the specific method of debugging the constructed PSO-BP model in S4 until the model parameters are optimal is:

[0084] A1. The PSO model has many parameters, and the main influencing parameters are learning rate, initial population size, maximum evolutionary generations, inertia weight, and learning factors C1 and C2; the main influencing parameter of the BP neural model is the number of hidden layer nodes. The PSO model is used to optimize the main influencing parameters of the BP neural network model, and other influencing parameters of the model are set based on experience.

[0085] A2. According to the preliminary experiment, the learning rate of the PSO-BP model was set to 0.001, 0.005, 0.01, and 0.05 respectively; the initial population size was set to 30, 40, and 50 respectively; the maximum evolutionary generations were set to 50, 60, 70, 80, and 90 respectively; the inertia weights were set to 0.6, 0.9, 1.2, and 1.5 respectively; C1 and C2 were set to 1.6 and 1.8, 1.6 and 2, 2 and 2, and 1.8 and 1.6 respectively; the number of hidden layer nodes was set to 5, 7, and 9 respectively; the training set data was input into the PSO-BP model for training to determine the optimal model parameters of the PSO and BP neural networks.

[0086] Specifically, Matlab 2021b is used to run the PSO-BP model, and the values ​​of the above-mentioned main influencing parameters are set in sequence in the PSO model and the BP neural model respectively. When setting or adjusting a main influencing parameter, the values ​​of the other main influencing parameters are fixed and not adjusted. When a main influencing parameter is optimized, the model accuracy during the adjustment process of the main influencing parameter is recorded, and the optimal PSO-BP model for the current main influencing parameter adjustment is determined according to the model accuracy. After obtaining the optimal PSO-BP model for this main influencing parameter, the current main influencing parameter is fixed, and the next main influencing parameter is adjusted in sequence until the optimal PSO-BP model of each parameter is obtained. Based on the sequential adjustment of different main influencing parameters, the PSO-BP models with different main influencing parameter values ​​are sequentially constructed through the above-mentioned design of the BP neural network model structure, construction of the PSO optimization algorithm model and initialization of the model parameters of the PSO algorithm model. The training set data are respectively input into the PSO-BP model after the main influencing parameters are adjusted. During the training process, the sample data and labels in the training set data are used as the basis for updating. The sample data is input into the BP neural model to obtain the corresponding output prediction result. The value of the fitness function is obtained by outputting the prediction result and the label. The value of the fitness function in the PSO optimization algorithm model is used to guide the update of the weight and threshold of the feedback BP neural model until all the training sets are traversed to complete the training of the PSO-BP model in the process of adjusting the main influencing parameters, and the training of the PSO-BP model in the process of adjusting the main influencing parameters is carried out. By adjusting the above-mentioned different main influencing parameters in sequence, the prediction accuracy of the PSO-BP model is further improved from the level of the main influencing parameters or hyperparameters.

[0087] A3. In the process of training different PSO-BP models with different main influencing parameters, the root mean square error RMSE and determination coefficient R of the test set are used. 2 Indicators determine model accuracy;

[0088] A4, taking the four characteristic quantities of each node flow, each node pressure, regional pressure difference and average pressure of DMA partition as decision variables, taking the nonlinear correlation between output flow and the four characteristic quantities as the objective function, minimizing the error between the objective function and the BP neural network model as the optimization purpose, taking the true value of output flow as the target position, adjusting the algorithm hyperparameters, i.e. the above-mentioned main influencing parameters and the network parameters of the BP model, so that the predicted value of output flow of the PSO-BP model keeps approaching the true value of output flow;

[0089] A5. Taking the model verified by the test set and its final accuracy as the goal, determine whether the judgment condition for ending the loop is met according to the model accuracy in A3. The judgment condition includes stopping or reaching the maximum number of loops if the accuracy reaches the preset model accuracy within the maximum number of loops. If it is not met, recalculate the traffic prediction value based on the updated model parameters and repeat A2-A4; if it is met, output the optimal PSO-BP model with optimized hyperparameters and network parameters, predict the data to be predicted through the optimal PSO-BP model, and obtain the final traffic prediction value.

[0090] As some embodiments, the root mean square error RMSE and the determination coefficient R in A3 2 The specific method for judging the accuracy of the indicator judgment model is as follows:

[0091]

[0092] Where n is the number of data in the cross-validation set, z i and are the measured and predicted values ​​of the water supply network pressure respectively, and z is the average value of the measured values.

[0093] As some embodiments, the evaluation standard accuracy R of the pipe network anomaly detection model in S5 is:

[0094] R = (number of abnormal data found / number of all abnormal data) × 100%

[0095] The technical solution of the present invention has the following advantages:

[0096] 1. The PSO-BP model for identifying newly added leakage based on the pressure difference and average pressure of DMA partitions in the present invention takes into account the main factors affecting pipeline network leakage, including regional pressure difference and regional average pressure, to ensure that the designed model is closer to the actual pipeline network leakage situation.

[0097] 2. The present invention mainly uses pressure data to distinguish and identify leakage, which is suitable for the situation that my country's water supply network generally only installs flow meters at regional entrances and exits, while a large number of pressure monitoring points are arranged. It has the characteristics of low cost and high practicality. In addition, the present invention can be well matched with various prediction models. With the development of computers and deep learning, more excellent pipeline prediction models will further improve the final leakage identification results of the present invention.

[0098] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A water supply network leakage identification method based on particle swarm optimization BP neural network is characterized by: include: Obtain flow and pressure data of the water supply network; The water supply network is divided into zones, and pressure characteristics are obtained according to the zone division results and pressure data; Constructing a prediction model, wherein the prediction model includes an optimization algorithm model and a neural network model, the input of the deep learning model is a feature variable, wherein the feature variable includes node flow data, pressure data and pressure characteristics, and the output is an output flow prediction value; Based on flow data, pressure data and pressure characteristics, the deep learning model is adjusted and optimized to obtain the final prediction model; The water supply network data to be tested and the measured output flow value are obtained, and the water supply network data to be tested are identified through the final prediction model to obtain the flow prediction value of the water supply network to be tested, and the measured output flow value and the flow prediction value of the water supply network to be tested are judged to obtain the water supply network leakage identification result.

2. The method according to claim 1, characterized in that The water supply network is zoned by DMA zoning, and the pressure characteristics include regional pressure difference and average pressure, wherein the regional pressure difference is the difference between the average value of the inlet pressure gauge data and the average value of the outlet pressure gauge data in each divided area in the zoning result; the average pressure is the average value of all pressure gauge data in each divided area.

3. The method according to claim 1, characterized in that The neural network model adopts a BP neural network model, wherein the BP neural network model comprises an input layer, a plurality of hidden layers and an output layer connected in sequence, wherein the hidden layer comprises a plurality of hidden nodes.

4. The method according to claim 3, characterized in that The optimization algorithm model adopts the PSO optimization algorithm.

5. The method according to claim 1, characterized in that The process of building a predictive model includes: Constructing a neural network model, and initializing the network structure and network parameters of the neural network model; An optimization algorithm model is constructed, and individual dimensions in a population of the optimization algorithm model are matched with network parameters of a neural network model, wherein the individual dimensions in the population are equal to the sum of the number of weights and thresholds in the neural network model. Initializing model parameters in the optimization algorithm model; The objective function and fitness function of the optimization algorithm model are set, wherein the objective function is the error between the output of the neural network model and the actual value, and the fitness function is the mean square error of the output of the neural network model.

6. The method according to claim 5, characterized in that The optimization process of the prediction model includes: Integrate and label the flow data, pressure data, and pressure features to obtain training sets and test sets; Iteratively input the training set, and in each iteration process, calculate the fitness of individuals in the population in the optimization algorithm model; according to the fitness of individuals in the population, update the population, and bring the updated weights and thresholds into the neural network model for parameter update, and perform the next iteration process based on the neural network model with updated parameters until the maximum number of iterations is reached, and the optimal solution of the population individuals is obtained, and the optimal solution of the population individuals is used as the initial value of the weight and threshold in the neural network model to obtain a pre-trained model; The neural network model is trained and tested through a training set and a test set to obtain an optimized prediction model.

7. The method according to claim 1, characterized in that The process of updating the population includes: According to the fitness of individuals in the population, the current individual extreme value and the current group extreme value are obtained, and the population is updated according to the current individual extreme value and the current group extreme value.

8. The method according to claim 6, characterized in that The process of tuning a prediction model includes: The influencing parameters of the optimization algorithm model and the topological structure of the neural network model are iteratively adjusted. In each iterative adjustment process, the prediction model is optimized, and the model accuracy of the optimized prediction model is judged. According to the model accuracy judgment result, the hyperparameters of the prediction model obtained in each iterative adjustment process are adjusted until the judgment condition for ending the iteration is met to obtain the final prediction model.

9. The method according to claim 8, characterized in that The accuracy judgment process includes: judging the model accuracy of the optimized prediction model through the root mean square error and the determination coefficient.

10. The method according to claim 8, characterized in that The process of judging the measured flow value and the predicted flow value of the water supply network to be measured includes: The difference calculation is performed on the measured flow value and the predicted flow value of the water supply network to be measured. If the difference calculation result is within the range of the model accuracy index and the normal leakage, it is considered that there is no abnormal leakage. Otherwise, there is an abnormal situation. The leakage point is detected and located according to the area where the flow meter corresponding to the measured flow value where the abnormal situation exists is located.

Citation Information

Patent Citations

  • Water pipe network leakage locating method and apparatus

    CN108596260A

  • Real-time leakage detection method, apparatus and system of water supply network and storage medium

    CN108984873A

  • Water supply pipe network leakage monitoring method

    CN109555979A

  • Method and device for determining position of leakage point in gas pipe network and medium

    CN115111537A

  • Water supply pipeline leakage detection method based on graph neural network

    CN115654381A