Pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm

Abstract

The invention discloses a pressure pipeline robot underground water supply pipeline leakage detection method based on a PSO-CNN-LSTM neural network algorithm, and the method comprises the steps: obtaining the monitoring data of a pressure pipeline robot, constructing a pipeline leakage data set, and dividing the pipeline leakage data set into a training set, a verification set and a test set according to a proportion; a PSO-CNN-LSTM model is constructed by using the pipeline leakage data set; a PSO-CNN-LSTM model is initialized, hyper-parameters are set, the training set is imported into the PSO-CNN-LSTM model, and model training is carried out; hyper-parameters are adjusted in sequence, a verification set data test model is introduced, change curves of loss values and accuracy rates of the model under different hyper-parameters are compared, and an optimal model is searched; and according to test set data, testing the optimal model, outputting various numerical evaluation indexes, and judging whether an expected value is reached or not. According to the invention, the leakage point in the pipeline can be accurately detected.

Description

Technical Field

[0001] The present invention relates to the technical field of underground water supply pipeline leakage detection, and in particular to a pressure pipeline robot underground water supply pipeline leakage detection method based on a PSO-CNN-LSTM neural network algorithm. Background Art

[0002] With the continuous development of the economy, the length of newly built water supply pipelines has exceeded 30,000 kilometers each year since 2008. By the beginning of 2024, the water supply coverage rate in most provinces in my country has exceeded 92%. Since water supply pipelines need to be buried underground and are pressure pipelines, they will be affected by internal water corrosion, external loads, and soil corrosion, so the probability of pipeline leakage is relatively high.

[0003] Traditional detection methods involve manual detection or repairing visible leaks only, while hidden leaks, which are more common in pipeline leaks, require active detection methods. In existing research, for example, Chinese patent CN111963909B, sensor nodes are arranged at each node of the underground pipeline network to obtain leakage noise signals, and then the leakage noise signals are processed to determine the location of the leakage point. However, the signal obtained by this method is mixed with too much noise, and the leakage point cannot be located, and the leakage point cannot be accurately detected in the pipeline. Summary of the invention

[0004] In order to solve the problem of underground water supply pipeline leakage detection, the present invention provides a pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] A pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm, comprising:

[0007] Obtain monitoring data of the pressure pipeline robot, construct a pipeline leakage dataset, and divide it into training set, validation set, and test set in proportion;

[0008] The PSO-CNN-LSTM model is constructed using the pipeline leakage dataset;

[0009] Initialize the PSO-CNN-LSTM model, set hyperparameters, import the training set into the PSO-CNN-LSTM model, and train the model;

[0010] Adjust the hyperparameters in order, introduce the validation set data to test the model, compare the loss value and accuracy curves of the model under different hyperparameters, and find the optimal model;

[0011] According to the test set data, the optimal model is tested, and various numerical evaluation indicators are output to determine whether they reach the expected values.

[0012] Further: the construction of the PSO-CNN-LSTM model using the pipeline leakage dataset includes:

[0013] Build a CNN-LSTM model;

[0014] Define the fitness function;

[0015] The hyperparameters of the CNN-LSTM model are used as optimization variables of the PSO algorithm, and the optimal hyperparameter combination is searched through the PSO algorithm;

[0016] Use the training set to train the CNN-LSTM model, and use the PSO algorithm to update the model's hyperparameters in each iteration;

[0017] The validation set is used to evaluate the performance of the model under different hyperparameters, and the best hyperparameter combination is selected to obtain the PSO-CNN-LSTM model.

[0018] Further: the construction of the CNN-LSTM model includes:

[0019] Use the CNN model to extract spatial features from the pipeline leakage dataset, perform pooling operations to reduce feature dimensions, and perform batch normalization;

[0020] Use the LSTM model to extract time series features from the pipeline leakage dataset and perform dimensionality reduction on the extracted time features;

[0021] Establish global interaction between spatial and temporal features, and generate enhanced spatiotemporal fusion features through weighted fusion;

[0022] The fused spatiotemporal features are subjected to deep convolution processing and feature decoupling and then fused.

[0023] Further: the establishment of global interaction between spatial features and temporal features and the generation of enhanced spatiotemporal fusion features by weighted fusion include:

[0024] Construct a feature self-attention module, input spatial features and temporal features into the module respectively, establish global feature associations in spatial and temporal dimensions through the self-attention mechanism, and calculate the similarity matrix between features;

[0025] Using the weighted fusion strategy, the spatial features and temporal features output by the feature self-attention module are weighted summed to generate fused features;

[0026] A multi-head attention mechanism is introduced to capture the interactive relationship between different subspace features and sub-temporal features of the fusion features through multiple independent attention heads to obtain enhanced fusion features.

[0027] Further: the deep convolution processing and feature decoupling and fusion of the fused spatiotemporal features include:

[0028] The fused spatiotemporal features are processed using a deep convolutional network;

[0029] Construct a feature decoupling module to decouple the fused features into spatial dominant features and temporal dominant features, and then fuse them again;

[0030] The fused features are input into the fully connected layer and the Softmax layer to perform classification or regression tasks and output the model prediction results.

[0031] Further: the step of adjusting the hyperparameters in sequence, introducing the validation set data to test the model, comparing the loss value and accuracy curves of the model under different hyperparameters, and finding the optimal model includes:

[0032] The hyperparameters include learning rate, total number of iterations, number of memory blocks to stack, number of mini-batches, and number of prediction steps.

[0033] Further: the optimal model is tested according to the test set data, and various numerical evaluation indicators are output to determine whether the expected value is achieved, including:

[0034] The evaluation indicators include accuracy, error curve, mean square error, and mean absolute error.

[0035] Compared with the prior art, the present invention has the following technical advances:

[0036] By combining CNN and LSTM, this application can effectively capture the spatial features (such as the leakage location and shape in the image) and temporal features (such as the time series changes of sound waves or pressure data) in pipeline leakage data. This fusion of spatiotemporal features can improve the model's understanding of complex patterns, thereby improving detection accuracy. PSO is a powerful global optimization method that can effectively search the hyperparameter space of the CNN-LSTM model. Through PSO optimization, the optimal hyperparameter combination can be found, the performance of the model can be enhanced, and the risk of overfitting or underfitting can be reduced.

[0037] This application adopts a phased tuning method, using validation set data to further tune hyperparameters after initial model training. By gradually adjusting key hyperparameters such as learning rate and number of iterations, the solution can find model settings that are more suitable for the current data set and ensure the best performance of the final model. Therefore, this application can accurately detect leaks in the pipeline. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0039] In the attached picture:

[0040] Figure 1 It is a flow chart of a pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm of the present invention;

[0041] Figure 2 It is a flow chart of constructing a PSO-CNN-LSTM model using a pipeline leakage data set in a pressure pipeline robot underground water supply pipeline leakage detection method based on a PSO-CNN-LSTM neural network algorithm of the present invention;

[0042] Figure 3 The present invention discloses a flow chart of constructing a CNN-LSTM model for a pressure pipeline robot underground water supply pipeline leakage detection method based on a PSO-CNN-LSTM neural network algorithm. DETAILED DESCRIPTION

[0043] The following examples describe the present invention in detail. In the description of the following examples, specific details such as specific system structures, technologies, etc. are proposed for illustration rather than limitation, so as to provide a thorough understanding of the present application embodiments. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details. In other cases, the detailed description of well-known systems, devices, circuits and methods is omitted to prevent unnecessary details from hindering the description of the present application.

[0044] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0045] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0046] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0047] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0048] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0049] The following will describe the embodiments of the present invention in conjunction with the accompanying drawings. The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0050] like Figure 1 As shown, a pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm includes:

[0051] Step 100: Acquire monitoring data of the pressure pipeline robot, construct a pipeline leakage dataset, and divide it into a training set, a validation set, and a test set in proportion.

[0052] Pipeline leakage signal collection method: The pressure pipeline robot enters the pipe and collects leakage data at the leakage location using the high-precision circular scanning sonar, pressure transmitter, hydrophone, positioning beacon, gyroscope, camera and other sensors it carries. The ratio of training set, validation set and test set is 5:1:1.

[0053] Combination Figure 2 , Step 200, construct a PSO-CNN-LSTM model using the pipeline leakage dataset.

[0054] In a possible implementation, step 200 may specifically include:

[0055] Combination Figure 3 , step 201, constructing a CNN-LSTM model;

[0056] Step 2011: Use the CNN model to extract spatial features from the pipeline leakage dataset, perform pooling operations to reduce feature dimensions, and perform batch normalization.

[0057] A multi-scale convolutional neural network (MultiScale CNN) is used to extract spatial features. Convolution kernels of different sizes are used to capture features at different levels of the image. This can enrich the diversity of spatial features. The multi-scale features extracted by CNN are pooled to reduce the feature dimension. At the same time, batch normalization technology is used to normalize the features and improve the stability of the features.

[0058] Step 2012: Use the LSTM model to extract time series features from the pipeline leakage dataset, and perform dimensionality reduction on the extracted time features.

[0059] Use a multi-layer long short-term memory network (Multilayer LSTM) to extract time series features. Through the multi-layer stacked LSTM structure, the long-term and short-term dependencies in the time features are captured layer by layer, and the extracted time features are reduced in dimension. For example, a linear layer (Fully Connected Layer) or principal component analysis (PCA) method is used to compress the time features to an appropriate dimension to prepare for subsequent fusion operations.

[0060] Step 2013: Establish global interaction between spatial features and temporal features, and generate enhanced spatiotemporal fusion features through weighted fusion.

[0061] A feature self-attention module is designed, spatial features and temporal features are input into the module respectively, global feature associations are established in spatial and temporal dimensions through the self-attention mechanism, and the similarity matrix between features is calculated to enhance the interaction between different features.

[0062] Assume that the spatial features extracted by the first two steps are Where H and W represent the height and width of the feature map, respectively, and C s The number of channels representing spatial features. The temporal features are Where T represents the time step, C t The number of channels representing the temporal features.

[0063] First, the spatial feature F sThree different matrices are generated through a layer of linear transformation: query matrix Q s , the bond matrix K s , and the value matrix V s :

[0064]

[0065] in, and is the learnable weight matrix.

[0066] Similarly, the time feature F t Generate Q by linear transformation t , K t , and V t :

[0067]

[0068] Calculate the similarity matrix S between spatial features s and the similarity matrix S between time features t :

[0069]

[0070] Among them, C s and C t is a scaling factor used to balance features of different dimensions.

[0071] Using the similarity matrix S s and S t For the corresponding value matrix V s and V t Perform weighted summation to obtain the enhanced spatial feature F s ' and enhanced temporal features F t ':

[0072] F s '=S s V s

[0073] F t '=S t V t

[0074] The enhanced spatial feature F s ' and time characteristics F t 'Combined by weighted fusion strategy, the generation formula of fusion features is as follows:

[0075] F fusion α·F s '+β·F t '

[0076] Among them, α and β are the learned weight parameters, which are used to dynamically adjust the weights of spatial features and temporal features in the fusion process.

[0077] For the fusion feature F fusion Regularization is performed, using batch normalization or layer normalization techniques to ensure the stability and consistency of the fused features in subsequent processing.

[0078] The fusion feature F fusion Divide into h independent subspaces, and perform attention calculations in each subspace separately;

[0079] The output results of all attention heads are concatenated together to form the final enhanced feature representation F enhanced :

[0080] In order to further enhance the feature expression capability, residual connections are added, followed by layer normalization:

[0081] F final =LayerNorm(F fusion +F enhanced )

[0082] Through the feature self-attention module, global feature associations are established in the spatial and temporal dimensions, and a weighted fusion strategy is used to generate fused features. Then, a multi-head attention mechanism is introduced to enhance the interaction and expression capabilities of spatiotemporal features, and finally a fused feature with powerful expression capabilities is generated.

[0083] Step 2014: perform deep convolution processing and feature decoupling and fusion on the fused spatiotemporal features.

[0084] Select a deep convolutional neural network (DeepCNN), such as ResNet or DenseNet, for further processing of the fused features.

[0085] The final fusion feature F final As input, it is input into a deep convolutional network, and the fused features are processed layer by layer through multiple convolutional layers and non-linear activation functions to extract higher-level spatiotemporal feature representations.

[0086] The output after the deep convolutional network is F deep ∈R H'×W'×C' , where H', W'' and C' are the height, width and number of channels of the processed feature map respectively. This feature representation is more abstract and can better characterize complex spatiotemporal relationships.

[0087] Design a feature decoupling module, the goal of which is to combine the fused features F after deep convolution deep Decoupling as a spatially dominant feature and time-dominant features The decoupling method can process spatial and temporal features respectively through two different branches. Each branch contains independent convolutional layers and pooling layers to extract features in spatial and temporal dimensions respectively.

[0088] Spatial branch processing F deep Get the spatial dominant feature:

[0089]

[0090] Time branch processing F deep Get the time-dominant feature:

[0091]

[0092] The decoupled spatial dominant features and time-dominant features To merge again:

[0093]

[0094] Among them, γ and δ are learnable weight parameters used to adjust the contribution of spatial and temporal features in the fusion process.

[0095] For the re-integrated feature F re-fusion Perform regularization, using batch normalization or layer normalization techniques, to ensure the stability of features in subsequent processing.

[0096] The re-integrated feature F re-fusion Perform a flattening operation and transform it into a one-dimensional vector f flatten ∈R N , where N represents the total dimension of the feature, and then the one-dimensional vector is input into the fully connected layer, which is calculated by the following formula:

[0097] f out =W fc f flatten +b fc

[0098] Among them, W fc and b fc are the weight matrix and bias vector of the fully connected layer respectively.

[0099] If it is a classification task, the output f of the fully connected layer out Input to the Softmax layer to calculate the probability distribution:

[0100]

[0101] If it is a regression task, the result of the fully connected layer is directly output as the predicted value

[0102] For example, to identify the specific location or type of leakage, the final output should be a classification result, that is, the probability distribution of each category is output through the Softmax layer, and the category with the highest probability is selected as the final prediction. For example, to predict the extent of leakage or some continuous variable, the final output should be a regression result, that is, directly output one or more continuous values ​​as the prediction result.

[0103] By performing deep convolution processing on the fused spatiotemporal features, designing a feature decoupling module to decouple and re-fuse the features, and finally inputting them into the fully connected layer and Softmax layer for classification or regression output, this process enhances the expressive power of spatiotemporal features and improves the overall performance of the model.

[0104] Step 202: define a fitness function;

[0105] In order to optimize the hyperparameters of the CNN-LSTM model, it is necessary to define a fitness function, such as the loss value of the model on the validation set.

[0106] Step 203: Use the hyperparameters of the CNN-LSTM model as optimization variables of the PSO algorithm, and search for the optimal hyperparameter combination through the PSO algorithm;

[0107] Write the code for the PSO algorithm, including initializing the population, calculating fitness, and updating the position of the particle swarm.

[0108] Step 204: Use the training set to train the CNN-LSTM model, and use the PSO algorithm to update the hyperparameters of the model in each iteration;

[0109] The hyperparameters of the CNN-LSTM model are used as optimization variables of the PSO algorithm, and the PSO algorithm is used to search for the optimal hyperparameter combination to optimize the performance of the model.

[0110] Step 205: Use the validation set to evaluate the performance of the model under different hyperparameters, and select the hyperparameter combination with the best performance to obtain the PSO-CNN-LSTM model.

[0111] The CNN-LSTM model is trained using the training dataset, and the PSO algorithm is used to update the model's hyperparameters in each iteration.

[0112] Step 300: Initialize the PSO-CNN-LSTM model, set hyperparameters, import the training set into the PSO-CNN-LSTM model, and train the model.

[0113] According to the constructed PSO-CNN-LSTM model, preprocessing (normalization, data cleaning) and so on are performed.

[0114] According to the constructed PSO-CNN-LSTM model, the data is initialized by normalization and data cleaning. The hyperparameters are set, and the pipeline leakage time series training set data is imported into the PSO-CNN-LSTM neural network model for model training.

[0115] Specifically, normalization is a common technique in data preprocessing, which aims to scale the data so that it falls into a small specific interval, which has the effect of accelerating training efficiency and improving detection accuracy. Data cleaning is a very important part of the data governance process. It refers to the cleaning, screening, deduplication, and formatting of data to ensure data quality and accuracy. Based on the high bandwidth and multi-threaded parallel computing characteristics of GPU and CPU, deep learning algorithms must be performed on high-performance computers. At the same time, in order to ensure the real-time prediction and the portability of the device, the PSO-CNN-LSTM model is run on the onboard Nvidia Jetson nano development board as much as possible.

[0116] Step 400: adjust the hyperparameters in sequence, introduce the validation set data to test the model, compare the change curves of the model loss value and accuracy under different hyperparameters, and find the optimal model.

[0117] According to the obtained model, the validation set data is introduced for testing. The hyperparameters are adjusted in sequence, and the change curves of the loss value and accuracy of the model under different hyperparameters are compared to find the optimal hyperparameters and determine the optimal model.

[0118] Specifically, the main hyperparameters are learning rate, total number of iterations, number of memory block stacks, number of mini-batches, number of prediction steps, etc.

[0119] The learning rate determines the speed of weight update. During model training, if the learning rate is too large, the model will easily cross the optimal value and lead to overfitting. If the learning rate is too small, the gradient descent process will be too slow. The learning rate needs to be set based on experience and continuous experiments.

[0120] The number of mini-batches determines the direction of the numerical gradient descent. If the number of mini-batches is too small, it is easy to cause large sample differences and difficulty in convergence. If the number of mini-batches is too large, the gradient direction will be basically stable, but it is easy to fall into the local optimal solution and reduce the accuracy.

[0121] Step 500: Test the optimal model based on the test set data, output various numerical evaluation indicators, and determine whether they reach the expected values.

[0122] According to the optimal model obtained, the test set data is introduced for testing, and various numerical evaluation indicators are output to determine whether they reach the expected values.

[0123] Specifically, the numerical evaluation indicators include accuracy, mean square error, and mean absolute error. The mean square error refers to the expected value of the square of the difference between the parameter estimate and the true value of the parameter, that is:

[0124]

[0125] The mean absolute error is the average of the absolute errors, that is:

[0126]

[0127] Control the pressure pipeline robot to re-enter the pipe for inspection, observe whether there is a leak at the predicted point, and determine whether the predicted result can meet the required requirements.

[0128] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0129] In various embodiments, the hardware implementation of the technology can directly adopt existing intelligent devices, including but not limited to industrial computers, PCs, smart phones, handheld stand-alone computers, floor-standing stand-alone computers, etc. Its input device preferably adopts a screen keyboard, its data storage and calculation module adopts existing memory, calculator, controller, its internal communication module adopts existing communication ports and protocols, and its remote communication adopts existing GPRS network, World Wide Internet, etc.

[0130] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0131] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0132] Each functional unit in each embodiment of the present invention may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit. If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and may also instruct the relevant hardware to complete it through a computer program. The computer program may be stored in a computer-readable storage medium, and the computer program may be executed by the processor to implement the steps of the above-mentioned method embodiments. Among them, the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electrical carrier signals and telecommunications signals.

[0133] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of protection of the claims of the present invention.

Claims

1. A pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm, characterized in that: include: Obtain monitoring data of the pressure pipeline robot, construct a pipeline leakage dataset, and divide it into training set, validation set, and test set in proportion; The PSO-CNN-LSTM model is constructed using the pipeline leakage dataset; Initialize the PSO-CNN-LSTM model, set hyperparameters, import the training set into the PSO-CNN-LSTM model, and train the model; Adjust the hyperparameters in order, introduce the validation set data to test the model, compare the loss value and accuracy curves of the model under different hyperparameters, and find the optimal model; According to the test set data, the optimal model is tested, and various numerical evaluation indicators are output to determine whether they reach the expected values.

2. According to claim 1, a pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm is characterized in that: The method of constructing a PSO-CNN-LSTM model using a pipeline leakage dataset includes: Build a CNN-LSTM model; Define the fitness function; The hyperparameters of the CNN-LSTM model are used as optimization variables of the PSO algorithm, and the optimal hyperparameter combination is searched through the PSO algorithm; Use the training set to train the CNN-LSTM model, and use the PSO algorithm to update the model's hyperparameters in each iteration; The validation set is used to evaluate the performance of the model under different hyperparameters, and the best hyperparameter combination is selected to obtain the PSO-CNN-LSTM model.

3. According to claim 2, a pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm is characterized in that: The construction of the CNN-LSTM model includes: Use the CNN model to extract spatial features from the pipeline leakage dataset, perform pooling operations to reduce feature dimensions, and perform batch normalization; Use the LSTM model to extract time series features from the pipeline leakage dataset and perform dimensionality reduction on the extracted time features; Establish global interaction between spatial and temporal features, and generate enhanced spatiotemporal fusion features through weighted fusion; The fused spatiotemporal features are subjected to deep convolution processing and feature decoupling and then fused.

4. According to claim 3, a pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm is characterized in that: The global interaction of the spatial features and the temporal features is established to generate enhanced spatiotemporal fusion features through weighted fusion, including: Construct a feature self-attention module, input spatial features and temporal features into the module respectively, establish global feature associations in spatial and temporal dimensions through the self-attention mechanism, and calculate the similarity matrix between features; Using the weighted fusion strategy, the spatial features and temporal features output by the feature self-attention module are weighted summed to generate fused features; A multi-head attention mechanism is introduced to capture the interactive relationship between different subspace features and sub-temporal features of the fusion features through multiple independent attention heads to obtain enhanced fusion features.

5. According to claim 4, a pressure pipeline robot underground water supply pipeline leakage detection method based on PSO-CNN-LSTM neural network algorithm is characterized in that: The deep convolution processing and feature decoupling and fusion of the fused spatiotemporal features include: The fused spatiotemporal features are processed using a deep convolutional network; Construct a feature decoupling module to decouple the fused features into spatial dominant features and temporal dominant features, and then fuse them again; The fused features are input into the fully connected layer and the Softmax layer to perform classification or regression tasks and output the model prediction results.

6. The method for detecting underground water supply pipeline leakage using a pressure pipeline robot based on a PSO-CNN-LSTM neural network algorithm according to claim 5 is characterized in that: The method of adjusting the hyperparameters in sequence, introducing the validation set data to test the model, comparing the loss value and accuracy curve of the model under different hyperparameters, and finding the optimal model includes: The hyperparameters include learning rate, total number of iterations, number of memory blocks to stack, number of mini-batches, and number of prediction steps.

7. The method for detecting underground water supply pipeline leakage using a pressure pipeline robot based on a PSO-CNN-LSTM neural network algorithm according to claim 6 is characterized in that: The optimal model is tested based on the test set data, and various numerical evaluation indicators are output to determine whether the expected values ​​are achieved, including: The evaluation indicators include accuracy, error curve, mean square error, and mean absolute error.

Images