Method, system, medium and equipment for improving leakage identification performance

Abstract

The invention relates to the field of water supply network leakage identification, and discloses a method, a system, a medium and equipment for improving leakage identification performance, and the method comprises the steps: carrying out the data processing of unlabeled monitoring data, dividing the processed unlabeled monitoring data into a homologous data pair and a non-homologous data pair, inputting the two data pairs into a pre-training model, and carrying out the feature extraction and pre-training; constructing a leakage detection and identification model based on the features extracted from the pre-training model, and migrating the data difference distinguishing capability learned from the unlabeled data to the leakage detection and identification model; the leakage detection and identification model is finely adjusted by adopting the labeled data, and the parameters of the leakage detection and identification model are continuously optimized through back propagation, so that the leakage identification performance is improved. According to the method, the problems of high data marking cost and high difficulty are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of water supply network leakage identification, and in particular to a method, system, medium and equipment for improving leakage identification performance. Background Art

[0002] The water supply network is an indispensable infrastructure for cities, shouldering the core responsibility of delivering clean and safe water resources. However, with the increasing problems of pipeline aging, corrosion, external damage and poor management, the leakage problem of the water supply network has gradually become a major obstacle to the efficient use and sustainable management of urban water resources. In order to effectively meet this challenge, leakage identification has become a key link in leakage management. Through advanced technical means, early detection and positioning of leaks in the water supply network can not only significantly reduce the leakage rate, but also provide strong support for the construction of a water-saving society, thereby better ensuring the long-term sustainable use of water resources.

[0003] Using machine learning techniques to automatically detect leak signals in monitoring data is a promising and efficient leak identification method. In recent years, leak identification methods based on deep learning have received widespread attention. However, these methods are usually supervised and require a large amount of labeled data (i.e. data that clearly indicates whether a leak exists) for model training. Obtaining labeled data requires on-site verification by professionals, and often requires mining and backfilling, which leads to high costs for building accurate and practical supervised leak identification models. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a method, system, medium and device for improving leakage identification performance, which effectively solves the problem of high cost and difficulty in labeling data.

[0005] To achieve the above-mentioned objectives, in the first aspect, the technical solution adopted by the present invention is: a method for improving leakage identification performance, which comprises: dividing the unlabeled monitoring data into homologous data pairs and non-homologous data pairs after data processing, and inputting both data pairs into a pre-training model for feature extraction and pre-training; constructing a leakage detection and identification model based on the features extracted from the pre-training model, and migrating the ability to distinguish data differences learned from the unlabeled data to the leakage detection and identification model; using labeled data to fine-tune the leakage detection and identification model, and continuously optimizing the parameters of the leakage detection and identification model through back propagation to improve the leakage identification performance.

[0006] Furthermore, the unlabeled monitoring data are divided into homologous data pairs and non-homologous data pairs after data processing, including: using a data augmentation method to generate two different versions for each unlabeled monitoring data time series, thereby forming homologous data pairs and non-homologous data pairs.

[0007] Furthermore, both data pairs are input into the pre-trained model for feature extraction, including: inputting the generated homologous data pairs and non-homologous data pairs into the feature extractor in the pre-trained model to extract low-dimensional features of the time series, and then converting the low-dimensional features into feature vectors through the mapping layer of the pre-trained model to calculate the contrast loss.

[0008] Furthermore, both data pairs are input into the pre-training model for pre-training, including: The performance of the pre-training process is evaluated using the contrast loss function, and the parameters of the pre-training model are continuously optimized through back-propagation; Among them, the goal of the contrast loss function is to maximize the similarity between the two data in each homologous data pair, and minimize the similarity between the two data in each non-homologous data pair, so that the feature extractor has the ability to distinguish data differences.

[0009] Further, the similarity is: quantified using the cosine similarity metric.

[0010] Furthermore, a leak detection and recognition model is constructed based on the features extracted from the pre-trained model, and the ability to distinguish data differences learned from unlabeled data is transferred to the leak detection and recognition model, including: The fine-tuning technology is used to transfer the ability of the feature extractor to distinguish data differences learned from unlabeled data to the leakage detection and recognition model. During the training process of the leakage detection and recognition model, the pre-trained parameters are used as the initial parameters of the feature extractor, while the parameters of the classification layer are randomly initialized.

[0011] Furthermore, the leak detection and recognition model is fine-tuned using the labeled data, including: using the labeled data to train the leak detection and recognition model, and using the cross-entropy loss function to evaluate the fine-tuning process, and continuously optimizing the parameters of the leak detection and recognition model through back propagation.

[0012] In a second aspect, the technical solution adopted by the present invention is: a system for improving leakage identification performance, comprising: The pre-training module divides the unlabeled monitoring data into homologous data pairs and non-homologous data pairs after data processing, and inputs both data pairs into the pre-training model for feature extraction and pre-training; The knowledge transfer module builds a leak detection and recognition model based on the features extracted from the pre-trained model, and transfers the ability to distinguish data differences learned from unlabeled data to the leak detection and recognition model; The model fine-tuning module uses labeled data to train the leak detection and recognition model, and continuously optimizes the parameters of the leak detection and recognition model through back propagation to improve the leak recognition performance.

[0013] In a third aspect, the technical solution adopted by the present invention is: a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by a computing device, the computing device executes any one of the above methods.

[0014] In a fourth aspect, the technical solution adopted by the present invention is: a computing device, comprising: one or more processors, a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any of the above methods.

[0015] The present invention adopts the above technical solution, which has the following advantages: The present invention breaks through the traditional development ideas of supervised leakage identification models and constructs a semi-supervised leakage identification model development method with three stages: pre-training, knowledge transfer, and fine-tuning. In the pre-training stage, unlabeled data is used, and its goal is to train a feature extractor that can distinguish data differences; in the knowledge transfer stage, a leakage detection and identification model is built based on the feature extractor in the pre-training model, and the ability to distinguish data differences learned from unlabeled data is transferred to the leakage detection and identification model; in the fine-tuning stage, labeled data is used, and its goal is to use fine-tuning technology to make the leakage detection and identification model quickly applicable to leakage identification tasks in real pipe networks. It effectively solves the problem of poor performance of leakage identification models under the conditions of high cost and difficulty of labeling data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a method for improving leakage identification performance in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In view of the high cost and difficulty of obtaining labeled data for traditional supervised models, the present invention proposes a method, system, medium and device for improving leak identification performance, which has low data requirements while still achieving competitive identification performance. In water supply networks, the scale of raw monitoring data (i.e., unlabeled data) that has not been verified by professionals is huge and can be obtained after the monitoring equipment is installed. By using these unlabeled data to build a model, the cost of model development will be greatly reduced.

[0018] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0019] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0020] In one embodiment of the present invention, a method, system, medium and device for improving leakage identification performance are provided. In this embodiment, Figure 1 As shown, the method comprises the following steps: 1) Pre-training stage: After data processing, the unlabeled detection data is divided into homologous data pairs and non-homologous data pairs, and both data pairs are input into the pre-training model for feature extraction and pre-training; 2) Knowledge transfer stage: Build a leak detection and recognition model based on the features extracted from the pre-trained model, and transfer the ability to distinguish data differences learned from unlabeled data to the leak detection and recognition model; 3) Model fine-tuning stage: Use labeled data to fine-tune the leak detection and recognition model, and continuously optimize the parameters of the leak detection and recognition model through back propagation to improve the leak recognition performance.

[0021] In the above step 1), the unlabeled monitoring data is divided into homologous data pairs and non-homologous data pairs after data processing, specifically: using the data augmentation method, two different versions are generated for each unlabeled monitoring data time series, thereby forming homologous data pairs and non-homologous data pairs. Among them, the homologous data pairs are composed of two amplified versions of the same time series, and the non-homologous data pairs are composed of two amplified versions of different time series.

[0022] In this embodiment, the data augmentation method may be, for example, adding white noise to the time series, randomly adjusting the data arrangement order, and the like.

[0023] In the above step 1), both data pairs are input into the pre-trained model for feature extraction, specifically: The generated homologous data pairs and non-homologous data pairs are input into the feature extractor in the pre-trained model to extract the low-dimensional features of the time series, and then the low-dimensional features are converted into feature vectors through the mapping layer of the pre-trained model to calculate the contrast loss.

[0024] In this embodiment, both data pairs are input into the pre-training model for pre-training, specifically: the pre-training process is evaluated using a contrast loss function, and the parameters of the pre-training model are continuously optimized through back propagation.

[0025] Among them, the goal of the contrast loss function is to maximize the similarity between the two data in each homologous data pair, and minimize the similarity between the two data in each non-homologous data pair, so that the feature extractor has the ability to distinguish data differences.

[0026] In this embodiment, the similarity is quantified using a cosine similarity index.

[0027] In the above step 2), the pre-trained feature extractor is directly connected to the classification layer to build a leakage detection recognition model.

[0028] In the above step 2), a leak detection and recognition model is constructed based on the features extracted from the pre-trained model, and the ability to distinguish data differences learned from unlabeled data is transferred to the leak detection and recognition model. Specifically, the ability to distinguish data differences learned by the feature extractor from unlabeled data is transferred to the leak detection and recognition model using fine-tuning technology; wherein: in the process of leak detection and recognition model training, the pre-trained parameters are used as the initial parameters of the feature extractor, while the parameters of the classification layer are randomly initialized. This is significantly different from the traditional supervised learning method that randomly initializes all parameters.

[0029] In the above step 3), the leak detection and recognition model is fine-tuned using the labeled data. Specifically, the leak detection and recognition model is trained using the labeled data, and the model fine-tuning process is evaluated using the cross entropy loss function.

[0030] In the above embodiments, the unlabeled and labeled data: the data source is the same, and the monitoring equipment is of the same type, such as a flow meter in a water supply network, or a noise monitor; but the latter has been verified by professionals to determine whether it represents a leak and can be used for training supervised models.

[0031] In the above embodiments, the feature extractor in the pre-trained model and the leakage detection and recognition model can be a neural network of any structure, such as a convolutional neural network, a recurrent neural network, etc. The model structure can be adjusted according to the input data. If it is a time series, it is preferably a recurrent neural network, and if it is a time-frequency spectrum, it is preferably a convolutional neural network.

[0032] In the above embodiments, the mapping of the pre-trained model and the classification layer in the leakage detection and recognition model are composed of a simple fully connected neural network, for example, may include two fully connected layers.

[0033] In one embodiment of the present invention, a system for improving leakage identification performance is provided, comprising: The pre-training module divides the unlabeled monitoring data into homologous data pairs and non-homologous data pairs after data processing, and inputs both data pairs into the pre-training model for feature extraction and pre-training; The knowledge transfer module builds a leak detection and recognition model based on the features extracted from the pre-trained model, and transfers the ability to distinguish data differences learned from unlabeled data to the leak detection and recognition model; The model fine-tuning module uses labeled data to fine-tune the leak detection and identification model, and continuously optimizes the parameters of the leak detection and identification model through back propagation to improve the leak identification performance.

[0034] In the above embodiment, the unlabeled monitoring data is divided into homologous data pairs and non-homologous data pairs after data processing. Specifically, a data augmentation method is used to generate two different versions for each unlabeled monitoring data time series, thereby forming homologous data pairs and non-homologous data pairs.

[0035] In the above embodiment, both data pairs are input into the pre-trained model for feature extraction, specifically: the generated homologous data pairs and non-homologous data pairs are input into the feature extractor in the pre-trained model to extract low-dimensional features of the time series, and then the low-dimensional features are converted into feature vectors through the mapping layer of the pre-trained model to calculate the contrast loss.

[0036] In this embodiment, both data pairs are input into the pre-training model for pre-training, specifically: the pre-training process is evaluated using a contrast loss function, and the parameters of the pre-training model are continuously optimized through back propagation; wherein the goal of the contrast loss function is to maximize the similarity between the two data in each homologous data pair, and minimize the similarity between the two data in each non-homologous data pair, so that the feature extractor has the ability to distinguish data differences.

[0037] Among them, the similarity is: quantified using the cosine similarity indicator.

[0038] In the above embodiment, a leakage detection and recognition model is constructed based on the features extracted from the pre-trained model, and the ability to distinguish data differences learned from unlabeled data is transferred to the leakage detection and recognition model, including: Use fine-tuning technology to transfer the ability of the feature extractor to distinguish data differences learned from unlabeled data to the leakage detection and recognition model; Among them, during the training process of the leakage detection and recognition model, the pre-trained parameters are used as the initial parameters of the feature extractor, while the parameters of the classification layer are randomly initialized.

[0039] In the above embodiment, the leak detection and recognition model is fine-tuned using the labeled data, including: using the labeled data to train the leak detection and recognition model, and using the cross entropy loss function to evaluate the knowledge transfer process.

[0040] The system provided in this embodiment is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for specific processes and detailed contents, which will not be repeated here.

[0041] In one embodiment of the present invention, a computing device is provided, which may be a terminal, and may include: a processor, a communication interface, a memory, a display screen, and an input device. Among them, the processor, the communication interface, and the memory communicate with each other through a communication bus. The processor is used to provide computing and control capabilities. The memory includes a non-volatile storage medium and an internal memory, and the non-volatile storage medium stores an operating system and a computer program. When the computer program is executed by the processor, the method in each of the above embodiments is implemented; the internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a management network, NFC (near field communication) or other technologies. The display screen may be a liquid crystal display screen or an electronic ink display screen, and the input device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computing device, or an external keyboard, touchpad or mouse, etc. The processor may call the logic instructions in the memory.

[0042] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0043] In one embodiment of the present invention, a computer program product is provided, wherein the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments.

[0044] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium stores server instructions, and the computer instructions enable a computer to execute the methods provided in the above embodiments.

[0045] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effect are similar to those of the above method embodiment, and will not be repeated here.

[0046] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0047] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0048] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for improving leakage identification performance, characterized in that: include: The unlabeled monitoring data is divided into homologous data pairs and non-homologous data pairs after data processing, and both data pairs are input into the pre-training model for feature extraction and pre-training; A leak detection and recognition model is built based on the features extracted from the pre-trained model, and the ability to distinguish data differences learned from unlabeled data is transferred to the leak detection and recognition model; The leak detection and identification model is fine-tuned using labeled data, and the parameters of the leak detection and identification model are continuously optimized through back propagation to improve the leak identification performance.

2. The method for improving leakage identification performance according to claim 1, characterized in that: After data processing, the unlabeled monitoring data are divided into homologous data pairs and non-homologous data pairs, including: A data augmentation method is used to generate two different versions for each unlabeled monitoring data time series, thus forming homologous data pairs and non-homologous data pairs.

3. The method for improving leakage identification performance according to claim 1, characterized in that: Both data pairs are input into the pre-trained model for feature extraction, including: The generated homologous data pairs and non-homologous data pairs are input into the feature extractor in the pre-trained model to extract the low-dimensional features of the time series, and then the low-dimensional features are converted into feature vectors through the mapping layer of the pre-trained model to calculate the contrast loss.

4. The method for improving leakage identification performance according to claim 3, characterized in that: Input both data pairs into the pre-trained model for pre-training, including: The performance of the pre-training process is evaluated using the contrast loss function, and the parameters of the pre-training model are continuously optimized through back-propagation; Among them, the goal of the contrast loss function is to maximize the similarity between the two data in each homologous data pair, and minimize the similarity between the two data in each non-homologous data pair, so that the feature extractor has the ability to distinguish data differences.

5. The method for improving leakage identification performance according to claim 4, characterized in that: The similarity is: quantified using the cosine similarity metric.

6. The method for improving leakage identification performance according to claim 1, characterized in that: A leak detection and recognition model is built based on the features extracted from the pre-trained model, and the ability to distinguish data differences learned from unlabeled data is transferred to the leak detection and recognition model, including: Use fine-tuning technology to transfer the ability of the feature extractor to distinguish data differences learned from unlabeled data to the leakage detection and recognition model; Among them, during the training process of the leakage detection and recognition model, the pre-trained parameters are used as the initial parameters of the feature extractor, while the parameters of the classification layer are randomly initialized.

7. The method for improving leakage identification performance according to claim 1, characterized in that: The leak detection and recognition model is fine-tuned using the labeled data, including: training the leak detection and recognition model using the labeled data, evaluating the fine-tuning process using the cross entropy loss function, and continuously optimizing the parameters of the leak detection and recognition model through back propagation.

8. A system for improving leak identification performance, characterized in that: include: The pre-training module divides the unlabeled monitoring data into homologous data pairs and non-homologous data pairs after data processing, and inputs both data pairs into the pre-training model for feature extraction and pre-training; The knowledge transfer module builds a leak detection and recognition model based on the features extracted from the pre-trained model, and transfers the ability to distinguish data differences learned from unlabeled data to the leak detection and recognition model; The model fine-tuning module uses labeled data to train the leak detection and recognition model, and continuously optimizes the parameters of the leak detection and recognition model through back propagation to improve the leak recognition performance.

9. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any one of the methods of claims 1 to 7.

10. A computing device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for executing any one of the methods described in claims 1 to 7.

Images