Construction method, device and equipment of pipe plugging prediction model, medium and product

Abstract

The embodiment of the invention provides a construction method and device of a pipe blockage prediction model, equipment, a medium and a product, and relates to the field of fault prediction, and the method breaks through the judgment limitation of a single pressure signal by introducing a first wet spraying machine multi-feature data set containing multi-dimensional operation parameters, provides comprehensive data support for judging the pipe blockage risk, and improves the pipe blockage prediction efficiency. Through collaborative design of a feature extraction module, a time sequence modeling module and an attention mechanism module in the hybrid model, local space-time features can be extracted from multi-dimensional data, a long-period time sequence dependency relationship is captured, key features related to pipe plugging are enhanced, deep mining of a pipe plugging precursor rule is achieved, and the accuracy of pipe plugging is improved. Compared with the prior art, early warning lag caused by judgment only depending on a pressure threshold value is avoided, meanwhile, hyper-parameters of the mixed model are optimized through a preset search algorithm, the model can adapt to different equipment working conditions, and through triple effects of multi-dimensional feature fusion, precursor rule accurate recognition and working condition self-adaptive adjustment, the pipe blockage prediction efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of fault prediction, and in particular to a method, apparatus, equipment, medium and product for constructing a pipe blockage prediction model. Background Technology

[0002] In wet spraying of concrete in tunnels, mines, and other engineering projects, the pumping system is prone to blockage due to factors such as concrete segregation and pipe wear under high pressure and continuous operation. Blockage can interrupt construction, damage equipment, and even cause safety accidents, severely impacting project efficiency and safety. Therefore, accurately predicting the risk of blockage is a key technical requirement for ensuring continuous and efficient construction and preventing malfunctions.

[0003] In existing technologies, the main oil pressure or pipeline pressure of the pumping system is monitored in real time, and the collected pressure signal is directly compared with a preset fixed safety threshold. When the pressure continuously exceeds the threshold, the system determines that there is a risk of pipe blockage and triggers an alarm.

[0004] However, existing technologies suffer from low efficiency in predicting pipe blockage. Prediction efficiency is limited by a fixed threshold, making it impossible to integrate multi-dimensional signals for comprehensive judgment. Furthermore, the threshold cannot be adaptively adjusted according to changes in equipment operating conditions and the environment, resulting in delayed warnings or high false negative rates, thus leading to low efficiency in predicting pipe blockage. Summary of the Invention

[0005] This application provides a method, apparatus, equipment, medium, and product for constructing a pipe blockage prediction model, in order to solve the problem of low efficiency in pipe blockage prediction in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for constructing a pipe blockage prediction model, comprising:

[0007] Obtain a first wet spraying machine multi-feature dataset; wherein, the first wet spraying machine multi-feature dataset is used to represent the operating status of the preset wet spraying machine pumping system within a preset first time period, and the end time of the first time period is earlier than the current time;

[0008] Based on the first wet spraying machine multi-feature dataset, a hybrid model is constructed, comprising a feature extraction module, a temporal modeling module, an attention mechanism module, and an output module. The feature extraction module performs local convolution and pooling on the first wet spraying machine multi-feature dataset to obtain local spatiotemporal features. The temporal modeling module models the dependencies of the local spatiotemporal features to obtain temporal dynamic features. The attention mechanism module evaluates the channel importance of the temporal dynamic features and performs adaptive weighting to obtain an attention-weighted feature vector. The output module determines whether pipe blockage occurs or not in a preset scenario based on the attention-weighted feature vector.

[0009] A pre-defined search algorithm is used to optimize multiple pre-defined hyperparameters in the hybrid model to obtain a pipe blockage prediction model. The multiple hyperparameters represent the structural configuration and training process of the hybrid model. The search algorithm is used to optimize the multiple hyperparameters so that the prediction performance of the hybrid model meets the pre-defined performance requirements. The pipe blockage prediction model is used to determine whether pipe blockage occurs or not in the scenario.

[0010] In one possible design, obtaining the first wet spraying machine multi-feature dataset includes:

[0011] Obtain the multi-feature dataset of the second wet spraying machine;

[0012] Anomaly detection is performed on the second wet spraying machine multi-feature dataset to obtain multiple anomalies; wherein, the multiple anomalies are data in the second wet spraying machine multi-feature dataset that deviate from the preset data distribution range;

[0013] The multiple outliers are removed from the second wet spraying machine multi-feature dataset to obtain the third wet spraying machine multi-feature dataset;

[0014] Missing values ​​were imputed in the third wet spraying machine multi-feature dataset to obtain the fourth wet spraying machine multi-feature dataset.

[0015] The data in the fourth wet spraying machine multi-feature dataset are normalized to obtain the first wet spraying machine multi-feature dataset.

[0016] In one possible design, after optimizing multiple preset hyperparameters in the hybrid model using a preset search algorithm to obtain the pipe blockage prediction model, the method further includes:

[0017] Obtain the multi-feature dataset of the fifth wet spraying machine; wherein, the multi-feature dataset of the fifth wet spraying machine is used to represent the operating status of the wet spraying machine pumping system at the current time;

[0018] The multi-feature dataset of the fifth wet spraying machine is input into the pipe blockage prediction model to obtain the pipe blockage prediction result; wherein, the pipe blockage prediction result is used to indicate whether pipe blockage occurs or not in the scenario.

[0019] In one possible design, inputting the multi-feature dataset of the fifth wet spraying machine into the pipe blockage prediction model to obtain the pipe blockage prediction result includes:

[0020] The multi-feature dataset of the fifth wet spraying machine is input into the feature extraction module to obtain the predicted local spatiotemporal features;

[0021] The predicted local spatiotemporal features are input into the temporal modeling module to obtain the predicted temporal dynamic features;

[0022] The predicted temporal dynamic features are input into the attention mechanism module to obtain the predicted attention weighted feature vector;

[0023] The predicted attention-weighted feature vector is input into the output module to obtain the pipe blockage prediction result.

[0024] In one possible design, inputting the multi-feature dataset of the fifth wet spraying machine into the feature extraction module to obtain predicted local spatiotemporal features includes:

[0025] The multi-feature dataset of the fifth wet spraying machine is subjected to two-level one-dimensional convolution processing to obtain preliminary convolution features;

[0026] The preliminary convolutional features are subjected to two-stage one-dimensional max pooling to obtain the predicted local spatiotemporal features.

[0027] In one possible design, inputting the predicted temporal dynamic features into the attention mechanism module to obtain the predicted attention-weighted feature vector includes:

[0028] The predicted time-series dynamic features are subjected to global average pooling to obtain a channel importance description vector; wherein, the predicted time-series dynamic features include multiple feature channels, and the channel importance description vector is used to represent the importance of each feature channel to the pipe blockage prediction;

[0029] The channel importance description vector is subjected to a nonlinear transformation to obtain the original attention weights; wherein, the original attention weights are preliminary weight values ​​obtained after performing a nonlinear transformation on the channel importance description vector;

[0030] The original attention weights are normalized to obtain the normalized channel attention weights;

[0031] The normalized channel attention weights are multiplied channel by channel by the predicted temporal dynamic features to obtain the attention-weighted feature vector.

[0032] Secondly, embodiments of this application provide an apparatus for constructing a pipe blockage prediction model, comprising:

[0033] The first acquisition module is used to acquire a first wet spraying machine multi-feature dataset; wherein, the first wet spraying machine multi-feature dataset is used to represent the operating status of the preset wet spraying machine pumping system within a preset first time period, and the end time of the first time period is earlier than the current time.

[0034] The first construction module is used to construct a hybrid model including a feature extraction module, a temporal modeling module, an attention mechanism module, and an output module based on the first wet spraying machine multi-feature dataset. The feature extraction module performs local convolution and pooling on the first wet spraying machine multi-feature dataset to obtain local spatiotemporal features. The temporal modeling module performs dependency modeling on the local spatiotemporal features to obtain temporal dynamic features. The attention mechanism module performs channel importance evaluation and adaptive weighting on the temporal dynamic features to obtain an attention-weighted feature vector. The output module determines whether pipe blockage or non-blockage occurs in a preset scenario based on the attention-weighted feature vector.

[0035] An optimization module is used to optimize multiple preset hyperparameters in the hybrid model using a preset search algorithm to obtain a pipe blockage prediction model; wherein, the multiple hyperparameters are used to represent the structural configuration and training process of the hybrid model, the search algorithm is used to optimize the multiple hyperparameters so that the prediction performance of the hybrid model meets preset performance requirements, and the pipe blockage prediction model is used to determine whether pipe blockage occurs or not in the scenario.

[0036] In one possible design, the first acquisition module includes:

[0037] The first acquisition unit is used to acquire the second wet spraying machine multi-feature dataset;

[0038] The detection unit is used to perform outlier detection on the second wet spraying machine multi-feature dataset and obtain multiple outliers; wherein, the multiple outliers are data in the second wet spraying machine multi-feature dataset that deviate from a preset data distribution range;

[0039] The elimination unit is used to eliminate the multiple outliers from the second wet spraying machine multi-feature dataset to obtain the third wet spraying machine multi-feature dataset.

[0040] The imputation unit is used to impute missing values ​​in the third wet spraying machine multi-feature dataset to obtain the fourth wet spraying machine multi-feature dataset.

[0041] The normalization unit is used to normalize the data in the fourth wet spraying machine multi-feature dataset to obtain the first wet spraying machine multi-feature dataset.

[0042] In one possible design, the apparatus for constructing the pipe blockage prediction model further includes:

[0043] The second acquisition module is used to acquire the multi-feature dataset of the fifth wet spraying machine; wherein, the multi-feature dataset of the fifth wet spraying machine is used to represent the operating status of the wet spraying machine pumping system at the current time;

[0044] The input module is used to input the multi-feature dataset of the fifth wet spraying machine into the pipe blockage prediction model to obtain the pipe blockage prediction result; wherein, the pipe blockage prediction result is used to indicate whether pipe blockage occurs or not in the scenario.

[0045] In one possible design, the input module includes:

[0046] The first input unit is used to input the multi-feature dataset of the fifth wet spraying machine into the feature extraction module to obtain the predicted local spatiotemporal features;

[0047] The second input unit is used to input the predicted local spatiotemporal features into the temporal modeling module to obtain the predicted temporal dynamic features;

[0048] The third input unit is used to input the predicted temporal dynamic features into the attention mechanism module to obtain the predicted attention weighted feature vector.

[0049] The fourth input unit is used to input the predicted attention-weighted feature vector into the output module to obtain the pipe blockage prediction result.

[0050] In one possible design, the first input unit includes:

[0051] The convolution component is used to perform two-level one-dimensional convolution processing on the multi-feature dataset of the fifth wet spraying machine to obtain preliminary convolution features.

[0052] A pooling component is used to perform two-stage one-dimensional max pooling on the initial convolutional features to obtain the predicted local spatiotemporal features.

[0053] In one possible design, the third input unit includes:

[0054] An average pooling component is used to perform global average pooling on the predicted time-series dynamic features to obtain a channel importance description vector; wherein, the predicted time-series dynamic features include multiple feature channels, and the channel importance description vector is used to represent the importance of each feature channel to the pipe blockage prediction.

[0055] A transformation component is used to perform a nonlinear transformation on the channel importance description vector to obtain the original attention weights; wherein, the original attention weights are preliminary weight values ​​obtained after performing a nonlinear transformation on the channel importance description vector;

[0056] A normalization component is used to normalize the original attention weights to obtain normalized channel attention weights.

[0057] The weighting component is used to multiply the normalized channel attention weights with the predicted temporal dynamic features channel by channel to obtain the attention-weighted feature vector.

[0058] Thirdly, embodiments of this application provide an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0059] The memory stores computer-executed instructions;

[0060] When the processor executes the computer execution instructions stored in the memory, it is used to implement the method for constructing the pipe blockage prediction model as described in any of the first aspects.

[0061] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method for constructing a pipe blockage prediction model as described in any of the first aspects.

[0062] Fifthly, embodiments of this application provide a computer program product, including a computer program, which, when executed by a processor, is used to implement the method for constructing a pipe blockage prediction model as described in any of the first aspects.

[0063] This application provides a method, apparatus, equipment, medium, and product for constructing a pipe blockage prediction model. By introducing a first wet spraying machine multi-feature dataset containing multi-dimensional operating parameters of the wet spraying machine pumping system, it overcomes the limitations of judging based on a single pressure signal, providing comprehensive data support for the comprehensive assessment of pipe blockage risk. The collaborative design of the feature extraction module, time series modeling module, and attention mechanism module in the hybrid model can extract local spatiotemporal features from multi-dimensional data, capture long-cycle time series dependencies, and strengthen key features related to pipe blockage, achieving in-depth mining of pipe blockage precursor patterns. This avoids the early warning lag caused by relying solely on pressure threshold judgment. At the same time, by optimizing the hyperparameters of the hybrid model through a preset search algorithm, the model can adapt to changes in different equipment operating conditions, environments, and material characteristics, replacing rigid fixed threshold settings. Finally, through the triple effect of multi-dimensional feature fusion, accurate identification of precursor patterns, and adaptive adjustment of operating conditions, the efficiency of pipe blockage prediction is improved. Attached Figure Description

[0064] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0065] Figure 1 A schematic diagram illustrating an application scenario of the method for constructing the pipe blockage prediction model provided in this application embodiment;

[0066] Figure 2 A flowchart illustrating the method for constructing the pipe blockage prediction model provided in this application embodiment. Figure 1 ;

[0067] Figure 3 A flowchart illustrating the method for constructing the pipe blockage prediction model provided in this application embodiment. Figure 2 ;

[0068] Figure 4 This application provides a technical roadmap for a method to predict pipe blockage in a wet spraying machine pumping system based on the SSA algorithm and optimized CNN-LSTM-SEnet.

[0069] Figure 5 This is a diagram of the CNN-LSTM-SEnet pipe blockage prediction network framework provided in the embodiments of this application;

[0070] Figure 6 A schematic diagram of the structure of the device for constructing the pipe blockage prediction model provided in the embodiments of this application;

[0071] Figure 7 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application.

[0072] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0073] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0074] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply difference. It should be noted that in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner. In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more.

[0075] It should be noted that the phrase "at...time" in the embodiments of this application can refer to the instant at which a certain situation occurs, or to a period of time after the occurrence of a certain situation; the embodiments of this application do not specifically limit this. Furthermore, the method, apparatus, device, medium, and product for constructing a pipe blockage prediction model provided in the embodiments of this application are merely examples; a method, apparatus, device, medium, and product for constructing a pipe blockage prediction model may also include more or fewer elements.

[0076] To facilitate a clear description of the technical solutions in the embodiments of this application, some terms and technologies involved in the embodiments of this application will be briefly introduced below:

[0077] The Sparrow Search Algorithm (SSA) is a swarm intelligence optimization algorithm that simulates the foraging and anti-predation behavior of sparrow populations in nature. This algorithm abstracts the solution to the optimization problem as the position of sparrows in the search space. By constructing a role model of discoverers, followers, and watchdogs and corresponding position update rules, it mimics the exploration, development, and early warning mechanisms of sparrow flocks in the process of finding food, thereby achieving efficient global search of the solution space of complex problems.

[0078] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0079] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0080] To clearly understand the technical solution of this application, the existing technical solutions will first be described in detail. Pipe blockage can interrupt construction, damage equipment, and even cause safety accidents, seriously affecting project efficiency and safety. Therefore, achieving accurate prediction of pipe blockage risks is a key technical requirement for ensuring continuous and efficient construction and preventing failures.

[0081] In existing technologies, the main oil pressure or pipeline pressure of the pumping system is monitored in real time, and the collected pressure signal is directly compared with a preset fixed safety threshold. When the pressure continuously exceeds the threshold, the system determines that there is a risk of pipe blockage and triggers an alarm. However, the prediction efficiency is limited by the fixed threshold, making it impossible to integrate multi-dimensional signals for comprehensive judgment. Furthermore, the threshold cannot be adaptively adjusted according to changes in equipment operating conditions and the environment, resulting in delayed warnings or a high rate of missed alarms. Therefore, existing technologies suffer from low efficiency in predicting pipe blockage.

[0082] Therefore, addressing the low efficiency of pipe blockage prediction in existing technologies, this research found that to solve this problem, multi-dimensional operational data from wet spraying pump systems can be integrated. A model with in-depth feature mining and adaptive operating conditions can replace the single pressure threshold judgment, accurately capturing early signs of pipe blockage and adapting to different operating scenarios, thus improving prediction efficiency: ① Multi-dimensional operational data from the pumping system can be collected and integrated to construct a comprehensive health status indicator system. By analyzing the correlation and synergistic change patterns among these indicators, a comprehensive judgment of pipe blockage risk can be achieved, improving the accuracy of early warnings. ② Multiple heterogeneous base models can be collaboratively constructed to build a discrimination mechanism. Prediction results from various models can be integrated through weighted voting and other methods. Simultaneously, multi-level risk level standards can be established, outputting differentiated early warning signals based on different risk levels to avoid delayed or excessive warnings. ③ Historical operational data including normal and blockage processes can be collected to train a classification or regression model. This model can automatically mine complex nonlinear features and interactions highly correlated with pipe blockage risk from massive multi-dimensional data and output risk probabilities or early warning levels.

[0083] Specifically, by continuously collecting multi-dimensional time-series data streams composed of various key operating parameters during the pumping process, machine learning models can be used to automatically learn and mine complex dynamic patterns and feature associations that characterize the precursors of pipe blockage from historical normal and abnormal data. Furthermore, the system can adaptively adjust its judgment logic and risk thresholds based on real-time equipment operating conditions and environmental changes, thereby achieving early, accurate, and dynamic prediction of pipe blockage risks.

[0084] This application provides a method, apparatus, equipment, medium, and product for constructing a pipe blockage prediction model. By introducing a first wet spraying machine multi-feature dataset containing multi-dimensional operating parameters of the wet spraying machine pumping system, it overcomes the limitations of judging based on a single pressure signal, providing comprehensive data support for the comprehensive assessment of pipe blockage risk. The collaborative design of the feature extraction module, time series modeling module, and attention mechanism module in the hybrid model can extract local spatiotemporal features from multi-dimensional data, capture long-term time series dependencies, and strengthen key features related to pipe blockage, achieving in-depth mining of pipe blockage precursor patterns. This avoids the early warning lag caused by relying solely on pressure threshold judgment. At the same time, by optimizing the hyperparameters of the hybrid model through a preset search algorithm, the model can adapt to changes in different equipment operating conditions, environments, and material characteristics, replacing rigid fixed threshold settings. Finally, through the triple effect of multi-dimensional feature fusion, accurate identification of precursor patterns, and adaptive adjustment of operating conditions, the efficiency of pipe blockage prediction is improved.

[0085] Based on the above-mentioned inventive discovery, the technical solution of this application is proposed.

[0086] The following describes the application scenarios of the method for constructing the pipe blockage prediction model provided in the embodiments of the present invention. Figure 1 This diagram illustrates an application scenario of the method for constructing the pipe blockage prediction model provided in this embodiment. Figure 1 As shown, the application scenario includes a mobile terminal 101 and a server 102. The mobile terminal 101 collects a multi-feature dataset of the first wet spraying machine and sends the multi-feature dataset of the first wet spraying machine to the server 102. Based on the multi-feature dataset of the first wet spraying machine, the server 102 constructs a hybrid model including a feature extraction module, a time series modeling module, an attention mechanism module, and an output module. The server 102 uses a preset search algorithm to optimize multiple preset hyperparameters in the hybrid model to obtain a pipe blockage prediction model.

[0087] The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0088] Figure 2 A flowchart illustrating the method for constructing the pipe blockage prediction model provided in this application embodiment. Figure 1 .like Figure 2 As shown, in this embodiment, the execution entity of this invention is a server. The method for constructing the pipe blockage prediction model provided in this embodiment includes the following steps:

[0089] S201. Obtain the first wet spraying machine multi-feature dataset; wherein, the first wet spraying machine multi-feature dataset is used to represent the operating status of the preset wet spraying machine pumping system within a preset first time period, and the end time of the first time period is earlier than the current time.

[0090] Specifically, the wet spraying machine pumping system can utilize sensors such as pressure sensors, flow sensors, frequency sensors, level sensors, and temperature sensors to collect data in real time, including main oil pressure, pipeline pressure, pumping frequency, concrete flow rate, accelerator dosage, hopper level, oil temperature, and air pressure, within a preset first time period. Simultaneously, the collected data is formatted and duplicate data is removed to form the first wet spraying machine multi-feature dataset. This step provides comprehensive and continuous operational status information of the wet spraying machine pumping system for the construction of the hybrid model, ensuring that the feature extraction module, temporal modeling module, and attention mechanism module can perform subsequent operations based on real and complete multi-dimensional operational data. This provides a reliable data foundation for mining pipe blockage-related features and model training.

[0091] S202. Based on the multi-feature dataset of the first wet spraying machine, a hybrid model is constructed, including a feature extraction module, a temporal modeling module, an attention mechanism module, and an output module. The feature extraction module is used to perform local convolution and pooling on the multi-feature dataset of the first wet spraying machine to obtain local spatiotemporal features. The temporal modeling module is used to perform dependency modeling on the local spatiotemporal features to obtain temporal dynamic features. The attention mechanism module is used to evaluate the channel importance and adaptively weight the temporal dynamic features to obtain attention-weighted feature vectors. The output module is used to determine whether pipe blockage or non-blockage occurs in the preset scenario based on the attention-weighted feature vectors.

[0092] Specifically, the preprocessed multi-feature dataset of the first wet spraying machine can be input into a pre-defined convolutional neural network (CNN) or residual network (ResNet) as a feature extraction module. Local spatiotemporal features are extracted from the data through local convolution and pooling operations. These local spatiotemporal features are then fed into a temporal modeling module composed of a Long Short-Term Memory (LSTM), a Gated Recurrent Unit (GRU), or a temporal Transformer to model temporal dependencies and capture long-cycle dynamic change features. Subsequently, the obtained temporal dynamic features are input into an attention mechanism module for channel importance evaluation and adaptive weighting. Finally, the attention-weighted feature vector is connected to the output module, which combines a fully connected layer with a Softmax activation function, forming a hybrid model that includes a feature extraction module, a temporal modeling module, an attention mechanism module, and an output module. This step is used to build a model architecture that can extract key information from multi-dimensional time-series data, capture dynamic change patterns, and output pipe blockage-related judgment results, providing core technical support for subsequent hyperparameter optimization and pipe blockage prediction.

[0093] S203. A preset search algorithm is used to optimize multiple preset hyperparameters in the hybrid model to obtain a pipe blockage prediction model. The multiple hyperparameters are used to represent the structural configuration and training process of the hybrid model. The search algorithm is used to optimize the multiple hyperparameters so that the prediction performance of the hybrid model meets the preset performance requirements. The pipe blockage prediction model is used to determine whether pipe blockage occurs or not in the scenario.

[0094] Specifically, the range of hyperparameters to be optimized in the hybrid model can be defined first. Hyperparameters can include kernel size, number of convolutional layers, number of recurrent units, number of attention heads, learning rate, batch size, and number of training epochs. Then, the Sparrow Search Algorithm (SSA) is selected as the preset search algorithm. One or more of accuracy, recall, precision, and F1 score are used as the criteria for judging model performance. The Sparrow Search Algorithm traverses the range of hyperparameters and iteratively verifies the running effect of the hybrid model under different combinations of hyperparameters. The optimal combination of hyperparameters that makes the model performance meet the preset requirements is selected. The optimal combination of hyperparameters is then substituted into the hybrid model to complete the configuration, resulting in the pipe blockage prediction model. This step is used to adapt the hybrid model to different equipment operating conditions, environments, and material characteristics, and output a prediction model with a stable structure and training process, providing an efficient and filtered technical carrier for judging the pipe blockage status of the wet spraying pumping system.

[0095] Among them, the pipe blockage prediction model can be applied to engineering construction fields related to wet concrete spraying construction, specifically covering tunnel engineering, mining engineering, underground engineering, water conservancy and hydropower engineering, and roadbed slope reinforcement engineering, etc. It is especially suitable for high-pressure and continuous operation environments where wet spraying machines are used for concrete spraying support.

[0096] The applications of pipe blockage prediction models are reflected in multi-dimensional engineering protection and efficiency improvement: First, risk prediction and safety protection: by capturing early warning signs of pipe blockage, it avoids construction interruptions, equipment overload damage, and even safety accidents such as concrete splashing and pipe bursts caused by pipe blockage, providing key protection for the personal safety of on-site construction personnel and the integrity of equipment; Second, construction efficiency optimization: reducing downtime for maintenance and pipe cleaning caused by pipe blockage, ensuring continuous wet spraying operations, and avoiding project delays, especially suitable for scenarios with high requirements for construction progress, such as tunnel excavation and mining development; Third, cost control: reducing the costs caused by pipe blockage. The system reduces concrete material waste and equipment maintenance costs, while also minimizing the additional manpower required for cleaning blocked pipes, indirectly lowering construction costs. Fourth, it empowers construction management by providing real-time feedback and risk warnings on the wet spraying machine's operating status to project managers, assisting in the development of construction plans and adjustments to operating parameters such as concrete mix proportions and pumping speed, thus promoting intelligent and refined management of the construction process. Fifth, it offers adaptability and generalization of applications, leveraging hyperparameter optimization to adapt to the wet spraying operation needs of different regions, environments, and material characteristics, providing stable and reliable pipe blockage prediction support for complex engineering scenarios.

[0097] This embodiment provides a method for constructing a pipe blockage prediction model. By introducing a first wet spraying machine multi-feature dataset containing multi-dimensional operating parameters of the wet spraying machine pumping system, it overcomes the limitations of judging based on a single pressure signal, providing comprehensive data support for the comprehensive assessment of pipe blockage risk. The collaborative design of the feature extraction module, temporal modeling module, and attention mechanism module in the hybrid model can extract local spatiotemporal features from multi-dimensional data, capture long-term temporal dependencies, and strengthen key features related to pipe blockage, achieving in-depth mining of pipe blockage precursor patterns. This avoids the early warning lag caused by relying solely on pressure threshold judgment. At the same time, by optimizing the hyperparameters of the hybrid model through a preset search algorithm, the model can adapt to changes in different equipment operating conditions, environments, and material characteristics, replacing rigid fixed threshold settings. Finally, through the triple effect of multi-dimensional feature fusion, accurate identification of precursor patterns, and adaptive adjustment of operating conditions, the efficiency of pipe blockage prediction is improved.

[0098] In one possible design, S201, obtain the first wet spraying machine multi-feature dataset, including:

[0099] S2011. Obtain the multi-feature dataset of the second wet spraying machine.

[0100] Specifically, pressure sensors, flow sensors, frequency sensors, level sensors, temperature sensors, and air pressure sensors can be deployed in key parts of the wet spraying machine's pumping system, such as the main oil circuit, pipelines, hopper, and pumping mechanism. Within a preset first time period, data such as main oil pressure, pipeline pressure, pumping frequency, concrete flow rate, accelerator dosage, hopper level, oil temperature, and air pressure can be collected in real time at fixed sampling intervals. The collected raw time-series data is then formatted and its integrity is verified to form a second wet spraying machine multi-feature dataset. This step provides raw multi-dimensional operating parameter data for subsequent outlier detection, missing value imputation, and normalization processing, providing initial data support for the construction of the first wet spraying machine multi-feature dataset.

[0101] The first wet spraying machine multi-feature dataset is a standardized dataset obtained by processing the second wet spraying machine multi-feature dataset after outlier detection and removal, missing value imputation, and normalization. It contains continuous, clean, and scale-consistent time-series data of various operating parameters of the wet spraying machine pumping system. The second wet spraying machine multi-feature dataset is the raw multi-dimensional time-series data collected in real time by various sensors deployed in the wet spraying machine pumping system within a preset time period. It has only undergone format unification and data integrity verification, without outlier and missing value processing or scale standardization. The core difference between the two lies in whether or not they have undergone preprocessing operations such as outlier removal, missing value imputation, and normalization. The purity, integrity, and scale consistency of the data differ.

[0102] S2012. Perform outlier detection on the multi-feature dataset of the second wet spraying machine to obtain multiple outliers; among them, the multiple outliers are data in the multi-feature dataset of the second wet spraying machine that deviate from the preset data distribution range.

[0103] Specifically, data distribution ranges can be set for each parameter in the multi-feature dataset of the second wet spraying machine, such as main oil pressure, pipeline pressure, and pumping frequency. Using box plot method or K-nearest neighbor anomaly detection method, values ​​of each parameter that exceed the corresponding distribution range are identified and marked, and multiple outliers are obtained. This step is used to accurately filter out invalid data in the multi-feature dataset of the second wet spraying machine that deviates from the normal operating state, providing a clear target for subsequent outlier removal operations and ensuring the accuracy and reliability of the dataset after subsequent processing.

[0104] S2013. Remove multiple outliers from the second wet spraying machine multi-feature dataset to obtain the third wet spraying machine multi-feature dataset.

[0105] Specifically, based on the labeling information obtained in the outlier detection phase, the specific location of each outlier in the second wet spraying machine multi-feature dataset can be located through data indexing. Values ​​marked as outliers are filtered and removed one by one according to parameter dimensions, and valid data that conforms to the preset data distribution range is retained to form the third wet spraying machine multi-feature dataset. This step is used to remove invalid data in the second wet spraying machine multi-feature dataset, reduce the interference of outlier data on subsequent missing value imputation, normalization processing and model construction, and ensure the effectiveness of the third wet spraying machine multi-feature dataset.

[0106] S2014. Impute missing values ​​in the third wet spraying machine multi-feature dataset to obtain the fourth wet spraying machine multi-feature dataset.

[0107] Specifically, we can first identify blank or invalid entries in the time series of each parameter in the multi-feature dataset of the third wet spraying machine. Based on the time series characteristics of the data and the correlation between parameters, we can select linear interpolation, K-nearest neighbor imputation, or prediction imputation based on long short-term memory networks to numerically supplement the identified missing entries, so that the time series data of each parameter is continuous and complete, and obtain the multi-feature dataset of the fourth wet spraying machine. This step is used to repair the incomplete information in the multi-feature dataset of the third wet spraying machine, eliminate the impact of missing data on subsequent processing and model construction, and provide continuous and complete multi-dimensional operating parameter data.

[0108] S2015. Normalize the data in the multi-feature dataset of the fourth wet spraying machine to obtain the multi-feature dataset of the first wet spraying machine.

[0109] Specifically, any one of Min-Max normalization, Z-Score normalization, or maximum absolute value normalization can be used to map the original values ​​of parameters such as main oil pressure, pipeline pressure, and pumping frequency in the multi-feature dataset of the fourth wet spraying machine to a preset unified value range, such as between 0 and 1. This eliminates the numerical deviation caused by the difference in magnitude of different parameters, resulting in the first multi-feature dataset of the wet spraying machine. This step is used to unify the data scale of multi-dimensional operating parameters, avoid the interference of parameter magnitude differences on subsequent feature extraction, time series modeling, and model training, and provide standardized data with consistent scale that can be directly used for calculation for the hybrid model.

[0110] The technical effect of this solution in this embodiment is that by systematically cleaning and standardizing the original data, including outlier removal, missing value imputation and normalization, a high-quality training dataset is constructed. This provides a key data foundation for the stable training and efficient learning of subsequent complex hybrid models, and enhances the robustness and generalization ability of the model.

[0111] Figure 3 A flowchart illustrating the method for constructing the pipe blockage prediction model provided in this application embodiment. Figure 2In this embodiment, in Figure 2 Based on the provided embodiments, the method for constructing the pipe blockage prediction model is further explained. The method for constructing the pipe blockage prediction model includes:

[0112] S301. Obtain the first wet spraying machine multi-feature dataset; wherein, the first wet spraying machine multi-feature dataset is used to represent the operating status of the preset wet spraying machine pumping system within a preset first time period, and the end time of the first time period is earlier than the current time.

[0113] S302. Based on the multi-feature dataset of the first wet spraying machine, construct a hybrid model including a feature extraction module, a temporal modeling module, an attention mechanism module, and an output module. The feature extraction module is used to perform local convolution and pooling on the multi-feature dataset of the first wet spraying machine to obtain local spatiotemporal features. The temporal modeling module is used to perform dependency modeling on the local spatiotemporal features to obtain temporal dynamic features. The attention mechanism module is used to evaluate the channel importance and adaptively weight the temporal dynamic features to obtain attention-weighted feature vectors. The output module is used to determine whether pipe blockage or non-blockage occurs in the preset scenario based on the attention-weighted feature vectors.

[0114] S303. A preset search algorithm is used to optimize multiple preset hyperparameters in the hybrid model to obtain a pipe blockage prediction model. The multiple hyperparameters are used to represent the structural configuration and training process of the hybrid model. The search algorithm is used to optimize the multiple hyperparameters so that the prediction performance of the hybrid model meets the preset performance requirements. The pipe blockage prediction model is used to determine whether pipe blockage occurs or not in the scenario.

[0115] S301-S303 are similar to S201-S203, and will not be described again in this embodiment.

[0116] S304. Obtain the multi-feature dataset of the fifth wet spraying machine; wherein, the multi-feature dataset of the fifth wet spraying machine is used to represent the operating status of the wet spraying machine pumping system at the current time.

[0117] Specifically, the pressure, flow, frequency, level, and temperature sensors already deployed in the wet spraying machine pumping system can be used to collect data such as main oil pressure, pipeline pressure, pumping frequency, concrete flow rate, accelerator dosage, hopper level, oil temperature, and air pressure during the real-time acquisition period corresponding to the current time, at sampling intervals consistent with the first wet spraying machine multi-feature dataset. Subsequently, the same outlier detection and removal, missing value imputation, and normalization processing as the second wet spraying machine multi-feature dataset are sequentially performed on this real-time time-series data to form the fifth wet spraying machine multi-feature dataset. This step is used to provide standardized multi-dimensional time-series data of the wet spraying machine pumping system under the current operating state, providing real-time data that meets the input requirements for the pipe blockage prediction model, and supporting the model to quickly output the corresponding pipe blockage prediction results.

[0118] The first wet spraying machine multi-feature dataset consists of operational status data of the wet spraying machine pumping system within a preset first time period where the cutoff time is earlier than the current time. It is obtained from the original second wet spraying machine multi-feature dataset after outlier detection and removal, missing value imputation, and normalization. This dataset is used to construct a hybrid model that includes modules such as feature extraction and temporal modeling. The fifth wet spraying machine multi-feature dataset contains operational status data of the wet spraying machine pumping system at the current time. It undergoes the same preprocessing as the second dataset, including outlier detection and removal, missing value imputation, and normalization, and is used as input to the optimized pipe blockage prediction model to obtain pipe blockage prediction results. The difference between the two datasets lies in their time dimension and purpose: the first dataset contains historical data, serving model construction and hyperparameter optimization; the fifth dataset contains real-time data, serving actual pipe blockage risk prediction. Although the preprocessing procedures are the same, the data acquisition time points are different—one is a historical time point, and the other is a current time point—corresponding to different operational scenarios.

[0119] S305. Input the multi-feature dataset of the fifth wet spraying machine into the pipe blockage prediction model to obtain the pipe blockage prediction result; wherein, the pipe blockage prediction result is used to indicate whether pipe blockage occurs or not in the scenario.

[0120] Specifically, the format and dimensions of the fifth wet spraying machine's multi-feature dataset can be adapted first to match the input requirements of the pipe blockage prediction model. Then, the adapted multi-feature dataset of the fifth wet spraying machine is input into the optimized pipe blockage prediction model. The model's internal feature extraction module, temporal modeling module, and attention mechanism module sequentially complete local spatiotemporal feature extraction, temporal dependency capture, and key feature weighting. Finally, the output module outputs the corresponding probability value. Based on the comparison between this probability value and the preset judgment threshold, the conclusion that pipe blockage has occurred or not in the scenario is determined, and the pipe blockage prediction result is obtained. This step is used to transform the standardized data of the current operating status of the wet spraying machine pumping system into clear pipe blockage-related judgment conclusions, providing a direct basis for judging the operating status of the wet spraying machine pumping system.

[0121] The technical effect of this solution in this embodiment is that it clarifies the specific steps for applying the trained pipe blockage prediction model to real-time scenarios. By acquiring the current operating status data and inputting it into the model for calculation, it achieves a seamless connection from model construction to actual early warning, completing a complete closed loop from offline training to online deployment of the technical solution, and ensuring the practicality and operability of the solution.

[0122] In one possible design, S305 inputs the multi-feature dataset of the fifth wet spraying machine into the pipe blockage prediction model to obtain the pipe blockage prediction results, including:

[0123] S3051. Input the multi-feature dataset of the fifth wet spraying machine into the feature extraction module to obtain the predicted local spatiotemporal features.

[0124] Specifically, the dimensions and format of the multi-feature dataset of the fifth wet spraying machine can be adjusted and adapted to match the input specifications of the feature extraction module. Then, the adapted dataset is input into the feature extraction module, where local features are captured through a preset convolution kernel. Pooling operations are used to compress the data dimensions while retaining key information, and finally, the corresponding data feature set is output to obtain the predicted local spatiotemporal features. This step is used to provide the temporal modeling module with the effective features that have been initially extracted, supporting the subsequent generation process of predicted temporal dynamic features.

[0125] Among them, predicting local spatiotemporal features involves adjusting the dimensions and adapting the format of the multi-feature dataset of the fifth wet spraying machine, capturing local correlation information of the data through the pre-set convolution kernel inside the module, and compressing the data dimensions while retaining key content through pooling operations. The final output is a set of data features containing local features and spatiotemporal distribution information of the operating parameters of the wet spraying machine pumping system. It can initially screen effective information in multi-dimensional real-time data and provide basic feature support for subsequent time series correlation analysis.

[0126] S3052. Input the predicted local spatiotemporal features into the time series modeling module to obtain the predicted time series dynamic features.

[0127] Specifically, the predicted local spatiotemporal features can first be adjusted in dimension and adapted to the temporal structure to meet the input requirements of the temporal modeling module. Then, the adapted predicted local spatiotemporal features are input into the temporal modeling module. The module captures the dynamic correlation and long-term change patterns of the features in the time dimension through the gating mechanism or self-attention mechanism inside the module, and outputs a feature vector containing temporal dependency information to obtain the predicted temporal dynamic features. This step is used to provide the attention mechanism module with core features carrying temporal dimension correlation information to support the accurate generation of the predicted attention-weighted feature vector.

[0128] Among them, predicting time-series dynamic features involves adjusting the dimensions of predicted local spatiotemporal features to adapt them to the time-series structure, and then capturing the dynamic correlation patterns and long-term change trends of features in the time dimension through the internal control mechanism or self-attention mechanism of the time-series modeling module. The final output is a feature vector carrying time-series dependent information, which can explore the evolution logic of multi-dimensional features over time and highlight the dynamic change characteristics of operating parameters.

[0129] S3053. Input the predicted temporal dynamic features into the attention mechanism module to obtain the predicted attention weighted feature vector.

[0130] Specifically, the predicted time-series dynamic features can first be adapted in terms of dimension and format to meet the input specifications of the attention mechanism module. Then, the adapted predicted time-series dynamic features are input into the module. The module assigns importance values ​​to different channels or spatiotemporal dimensions of the features through the internal weight calculation logic. The features are then adaptively weighted and fused according to the assignment results, and the output feature vector that enhances key information is obtained as the predicted attention weighted feature vector. This step is used to highlight the core information related to pipe blockage in the predicted time-series dynamic features and suppress irrelevant and redundant information, providing the output module with accurate and focused feature input.

[0131] Among them, the predictive attention-weighted feature vector is a feature vector that is generated by adapting the predicted time-series dynamic features to dimensions and formats, and then assigning importance values ​​to different channels or spatiotemporal dimensions of the features through the weight calculation logic inside the attention mechanism module. The feature vector is then adaptively weighted and fused according to the assignment results. The final output is a feature vector that strengthens core information and suppresses redundant information. It can focus on key features related to blockage and improve the targeting of subsequent judgment processes.

[0132] S3054. Input the predicted attention-weighted feature vector into the output module to obtain the pipe blockage prediction result.

[0133] Specifically, the predicted attention-weighted feature vector can first be adapted in terms of dimension and format to match the input specifications of the output module. Then, the adapted feature vector is passed into the output module, where it is mapped through a fully connected layer. The probability values ​​corresponding to the two cases of pipe blockage and non-blockage are calculated by the Softmax activation function. The probability values ​​are compared with the preset judgment threshold to determine and output the conclusion that pipe blockage or non-blockage has occurred in the scenario, thus obtaining the blockage prediction result. This step is used to transform the feature vector focusing on the core information related to pipe blockage into a clear judgment conclusion, providing a direct and executable basis for judging the operating status of the wet spraying pumping system.

[0134] The pipe blockage prediction result is obtained by adapting the predicted attention-weighted feature vector to the dimension and format, inputting it into the output module, completing feature mapping through a fully connected layer, calculating the corresponding probability values ​​for the two cases of pipe blockage and non-blockage through the Softmax activation function, and then comparing the probability values ​​with the preset judgment threshold. Finally, the output is a clear conclusion on whether pipe blockage has occurred or not in the scenario. It can directly reflect the pipe blockage related situation of the wet spraying pump system under the current operating state and provide an intuitive basis for judging the system operating status.

[0135] The technical effect of this solution in this embodiment is that by limiting the internal execution order and data flow path of the pipe blockage prediction model during real-time prediction, the complete and deterministic computational logic of the model from multiple feature inputs to the final result output is clarified, which enhances the interpretability and reproducibility of the technical solution and provides a clear structured foundation for further optimization or verification of the functions of each module.

[0136] In one possible design, S3051 inputs the multi-feature dataset of the fifth wet spraying machine into the feature extraction module to obtain the predicted local spatiotemporal features, including:

[0137] S30511. Perform two-level one-dimensional convolution processing on the multi-feature dataset of the fifth wet spraying machine to obtain preliminary convolution features.

[0138] Specifically, the dimensions and format of the fifth wet spraying machine's multi-feature dataset can be adjusted and adapted to meet the input requirements of two-stage one-dimensional convolution processing. The first-stage one-dimensional convolution uses a pre-sized kernel, a fixed stride, and padding operations to capture local features of the adapted dataset and enhance feature expression through activation functions. The output of the first-stage convolution is then fed into the second-stage one-dimensional convolution, which uses the same or different kernel parameters, stride, and padding operations as the first stage to further extract deep local correlation information. Finally, a feature set integrating the features of the two-stage convolution is output to obtain preliminary convolution features. This step is used to mine the local correlation information between different parameters in the fifth wet spraying machine's multi-feature dataset layer by layer, strengthen the effective feature expression in the data, and provide input data rich in local features for subsequent one-dimensional max pooling processing.

[0139] S30512. Perform two-stage one-dimensional max pooling on the initial convolutional features to obtain the predicted local spatiotemporal features.

[0140] Specifically, the initial convolutional features can be adapted in terms of format and dimension to meet the input requirements of two-stage one-dimensional max pooling. The first-stage one-dimensional max pooling uses a pooling window of a preset size and slides through the initial convolutional features with a fixed stride to extract the maximum value within each window as the feature representative of that window. The output of the first-stage pooling is then fed into the second-stage one-dimensional max pooling, which repeats the above operation with the same or different pooling window size and stride as the first stage to further compress the data dimension and retain the core local features. Finally, the feature set is output to obtain the predicted local spatiotemporal features. This step is used to reduce the data dimension and computational complexity of the initial convolutional features, retain the key local correlations and spatiotemporal distribution information in the data, and remove redundant content, providing efficient and focused feature input for the subsequent time series modeling module.

[0141] The technical effect of this scheme in this embodiment is that by using a specific structure of two-level one-dimensional convolution followed by two-level one-dimensional max pooling in the feature extraction module, efficient, hierarchical abstraction and dimensionality reduction of local abnormal patterns in multi-dimensional time series signals are achieved. This enhances the robustness and pertinence of the model in capturing key early fault features from complex raw data, and provides a clean and information-dense feature representation for subsequent time series modeling.

[0142] In one possible design, S3053 inputs the predicted temporal dynamic features into the attention mechanism module to obtain the predicted attention-weighted feature vector, including:

[0143] S30531. Perform global average pooling on the predicted time-series dynamic features to obtain the channel importance description vector; wherein, the predicted time-series dynamic features include multiple feature channels, and the channel importance description vector is used to represent the importance of each feature channel to the pipe blockage prediction.

[0144] Specifically, the predicted time-series dynamic features can first be adapted in terms of format and dimension to meet the processing requirements of global average pooling. For the multiple feature channels contained in the feature, the average value of all time-series position data under each channel is calculated. The average value corresponding to each channel is arranged in channel order to form a one-dimensional vector, which is used to obtain the channel importance description vector. This step is used to extract the global information of each feature channel, quantify the basic feature contribution of each channel, and provide standardized channel importance basic data for the subsequent calculation of the original attention weight.

[0145] S30532. Perform a nonlinear transformation on the channel importance description vector to obtain the original attention weights; wherein, the original attention weights are the preliminary weight values ​​obtained after performing a nonlinear transformation on the channel importance description vector.

[0146] Specifically, the channel importance description vector can first be dimension-adapted to conform to the input specifications of the nonlinear transformation. The adapted vector is then fed into a transformation network consisting of fully connected layers and activation functions. The vector is dimensionally compressed through the first-level fully connected layer, and a nonlinear feature mapping is introduced through the ReLU activation function. Finally, it is restored to the same dimension as the original number of feature channels through the second-level fully connected layer to obtain the original attention weights. This step is used to extract the differential contribution information of each channel in the channel importance description vector and generate preliminary weight data that can reflect the relative importance of the channels, providing a basis for subsequent normalization processing.

[0147] S30533. Normalize the original attention weights to obtain the normalized channel attention weights.

[0148] Specifically, the original attention weights can first be adapted in terms of format and dimension to meet the input requirements of normalization processing. The Softmax function is then used to calculate the adapted original attention weights. Through the probability mapping characteristics of the function, each weight value is transformed to between 0 and 1, and the sum of all weight values ​​is 1. The normalized weight sequence is then output to obtain the normalized channel attention weights. This step is used to unify the numerical scale of each channel weight, making the importance weights of different channels directly comparable. This provides a consistent and logically reasonable weight basis for the subsequent channel-by-channel multiplication operation with the predicted time-series dynamic features.

[0149] S30534. Multiply the normalized channel attention weights with the predicted temporal dynamic features channel by channel to obtain the attention-weighted feature vector.

[0150] Specifically, the channel dimension of the normalized channel attention weights and the predicted temporal dynamic features can be verified first to ensure that the number of channels is consistent and the data format is matched. Then, according to the feature channel order, all temporal data of each channel in the predicted temporal dynamic features are multiplied with the corresponding normalized channel attention weights. The feature information of each channel after weight modulation is retained, and the integrated feature vector is output to obtain the attention-weighted feature vector. This step is used to strengthen the information expression of important channels in the predicted temporal dynamic features through weight modulation, weaken the redundant information of irrelevant channels, generate vector data focusing on core features, and provide more targeted feature input for the output module.

[0151] The technical effect of this scheme in this embodiment is that by decomposing the channel attention mechanism into a calculation process of global pooling compression, nonlinear transformation to generate weights, normalization constraints, and channel-by-channel weighting, the internal working mechanism of adaptive evaluation of the model and strengthening of key feature channels and suppression of non-key channels is specified, thereby enhancing the interpretability and computational stability of the model feature selection process and ensuring the effectiveness of dynamic focusing on multi-source heterogeneous features.

[0152] Figure 4 This application provides a technical roadmap for a pipe blockage prediction method for a wet spraying machine pumping system based on SSA algorithm-optimized CNN-LSTM-SEnet. The specific process is as follows:

[0153] ① The collected multi-tunnel pumping dataset was preprocessed to remove abnormal noise data and perform normalization, and blockage judgment rules were set. ② A CNN-LSTM-SEnet hybrid prediction model framework was constructed. Multi-dimensional temporal features were extracted through convolutional layers, dynamic temporal dependencies were captured using a Long Short-Term Memory (LSTM) network, and key features were adaptively strengthened through the SEnet module. ③ The SSA algorithm was used to globally optimize the model's hyperparameters, and the prediction error was used as the fitness function to iteratively search for the optimal parameter combination. ④ The model was trained on the training and validation sets after temporal slices. The training process was dynamically terminated through an early stopping mechanism, and the model weights with the highest accuracy on the validation set were saved. ⑤ Finally, the trained CNN-LSTM-SEnet prediction model was deployed to the field for practical application, and real-time pipe blockage fault early warning was achieved based on the prediction parameters and set thresholds.

[0154] Figure 5 The CNN-LSTM-SEnet pipe clogging prediction network framework provided in the embodiments of this application is shown in the following diagram:

[0155] (1) Data preprocessing and pipe blockage judgment rules: ① Collect pumping parameter datasets for multi-tunnel pumping systems, including parameters such as main oil pump operation, air compressor operation, pumping speed (set value), pumping pressure, parking status, accelerator dosage (set value), vibration, and motor current; ② Clean and filter the pumping parameter data, use box plot method to detect and remove outliers based on interquartile range (IQR), and fill missing data using linear interpolation; ③ Apply Min-Max normalization to map multi-source heterogeneous pumping parameter data to the [0, 1] interval to eliminate dimensional differences; ④ Use sliding time window slicing technology to generate sample sequences, with a window length set to 120 sampling periods (corresponding to 120 seconds of actual pumping process) and a step size of 5 periods; ⑤ Establish pipe blockage judgment rules based on expert experience: when the pumping pressure continuously exceeds the rated pumping pressure threshold, it is judged as an abnormal pipe blockage state.

[0156] (2) Construct a CNN-LSTM-SEnet hybrid prediction model:

[0157] ① Use the cleaned training set pump parameter feature data as the input layer:

[0158]

[0159] in, For time step, Let X be the number of features, X be the multi-dimensional time-series pumping parameter feature data matrix of the input model, and R be the set of real numbers.

[0160] ② CNN Feature Extraction: A feature extraction network consisting of two one-dimensional convolutional layers (Conv1D+ReLU) and one max-pooling layer (MaxPool) is constructed. Multi-scale convolutional kernels capture spatiotemporal features through local receptive fields, while the max-pooling layer reduces dimensionality and suppresses overfitting through non-linear downsampling. The preprocessed normal operating condition data is input into the Convolutional Neural Network (CNN), and after feature extraction, it is input into an LSTM for temporal modeling. The specific formula is as follows:

[0161] The first convolutional layer is:

[0162]

[0163] The second convolutional layer is:

[0164]

[0165] The pooling layer is:

[0166]

[0167] in, , These are the weights of the first convolutional layer. The weight bias for the first convolutional layer, , The weights of the second convolutional layer, For the weight bias of the second convolutional layer, The number of convolution kernels, This refers to the window size for pooling operations, also known as the "pooling kernel size". For characteristic number, This is the output feature map of the first convolutional layer. This is the output feature map of the second convolutional layer. This is the output feature map of the pooling layer.

[0168] ③ Integrate LSTM, set the LSTM layer range to [1, 150], the neural memory unit range to [1, 256], and the dropout rate range to [0, 1], to capture the long and short-term temporal dependencies of the pumping parameters.

[0169] Let the SSA-optimized LSTM have the following parameters: number of layers L and number of neurons per layer N.

[0170]

[0171] in, (Number of floors) M represents the number of neurons per layer.

[0172] Dropout layer:

[0173]

[0174] Where D is the output of the Dropout layer after processing H. This is the dropout rate value optimized by SSA. rate is the dropout rate.

[0175] ④ Introduce the Squeeze-and-Excitation (SE) module from the SENet network to generate channel description vectors through global average pooling, and then generate channel attention weights through two fully connected layers.

[0176] (3) Optimize the hyperparameters of LSTM based on SSA algorithm: ① Initialize sparrow population and set the number of hidden layer nodes of LSTM (64-256), learning rate (1e-4-1e-2), and training rounds (50-300) as optimization dimensions.

[0177] ② Define the fitness function as a weighted combination of the validation set F1-score and mean squared error (weight ratio 7:3) to ensure classification accuracy and severity assessment precision; for each sparrow individual:

[0178]

[0179] Wherein, F1-score is the classification F1 value of the validation set (range [0, 1]), MSE is the mean squared error of the validation set (which needs to be standardized), and Fitness is the fitness value calculated by this formula, used to evaluate the performance of each individual sparrow.

[0180] ③ Sparrows are divided into discoverers, followers, and watchdogs, and their position update formulas are as follows:

[0181] Discoverer Update (Global Search):

[0182]

[0183] in, Indicates the first The middle generation The sparrow in the first The position of the dimension (i.e., the hyperparameter value). This represents the current iteration number. The maximum number of iterations, This is a constant, typically taking a value between (0, 1] (e.g., 0.01), used to control the step size decay. A random number in the range [0, 1] is used to simulate changes in the environment. As a safety threshold, it is set to 0.8 for this test. Let be a random number that follows a standard normal distribution (mean 0, variance 1). It is a unit vector. This is the index of the current sparrow.

[0184] Follower updates (local search):

[0185]

[0186] in, Let be the updated position of the i-th follower individual in the j-th dimension in the (t+1)-th generation. For the best position of the discoverer Randomly select elements The matrix, , Let j be the position of the individual with the worst fitness in the current population in the j-th dimension. Let represent the position of the best individual among the discoverers in the j-th dimension (i.e., the optimal position found by the current discoverer).

[0187] Vigilant Update (Escape Local Optimum):

[0188]

[0189] in, This represents the position of the best-fit individual in the current population in the j-th dimension. This is a step size control parameter, which typically follows a standard normal distribution (mean 0, variance 1). The fitness value of the current vigilant. This represents the current globally optimal fitness value. This represents the current worst-case fitness value globally. A random number uniformly distributed in the range [-1, 1] is used to control the direction of movement. It should be a very small constant (e.g., 1e-8) to avoid the denominator being zero.

[0190] Finally, the maximum number of iterations is set to 300. If the optimal fitness does not improve for 15 consecutive generations, the search is terminated early and the optimal solution set of hyperparameters is determined.

[0191] (4) Model training and validation: ① Five-fold time series cross-validation is used to ensure the temporal continuity of the training set, validation set, and test set; ② A dynamic class weight strategy is applied to adjust the weight of the loss function according to the real-time class distribution to alleviate the problem of sparse samples in the bottleneck; ③ An early stopping mechanism is used to set a patience value of 20. Training is terminated when the verification loss fails to decrease for 20 consecutive iterations.

[0192] (5) Model deployment and application: ① Integrate the fully trained model into the PC system of the wet spraying trolley; ② Collect various parameters during the spraying process in real time and dynamically input these data into the model according to the preset time interval; ③ The model predicts whether pipe blockage will occur within a specified time period by “establishing pipe blockage judgment rules based on expert experience”, providing timely warnings and decisions for operators.

[0193] A CNN-LSTM-SEnet neural network optimized based on SSA is used to predict pipe blockage faults in wet spraying machines. It integrates convolutional layers to extract spatiotemporal features, LSTM to capture the temporal dependencies of multiple features, and SENet channel attention weighting mechanism to reduce the risk of missed detections caused by single parameter dependence.

[0194] By integrating the SENet attention mechanism to adaptively enhance key features, channel description vectors are generated through global average pooling and feature weights are assigned through fully connected layers. This improves the model's sensitivity to key fault features in multi-source heterogeneous pumping parameters and suppresses non-critical or noise interference features.

[0195] SSA is used to dynamically optimize the hyperparameters of LSTM. The fitness function is a weighted combination of the validation set F1-score and mean squared error. The optimal solution set of hidden layer nodes, learning rate and training epochs is searched iteratively to solve the gradient vanishing / exploding problem and effectively improve the robustness and prediction accuracy of the model for fault prediction of wet spraying pumping system.

[0196] By deploying the pipe blockage prediction model on the PC of the wet spraying machine pumping system, data can be collected and analyzed in real time, which can predict the risk of impending pipe blockage and provide operators with real-time early warning and decision support.

[0197] The key points of using an SSA-optimized CNN-LSTM-SEnet neural network to predict pipe blockage faults in wet spraying machines are:

[0198] (1) Multi-dimensional time-series features advance multi-step early warning: By inputting multi-dimensional time-series features of wet spraying machine in parallel, a unified time-dimensional input sequence is constructed. Samples are generated using sliding time window technology, so that the model can learn the coupling relationship between multiple features at the same time and effectively capture the pipe blockage precursor signals caused by multiple potential factors.

[0199] (2) A strategy for incorporating SE attention mechanism: By integrating the SE module into the CNN-LSTM framework, an adaptive feature enhancement mechanism is formed: global average pooling is used to generate channel description vectors, and feature weights are assigned through fully connected layers to enhance the discriminativeness of key features and suppress non-key or noisy features.

[0200] (3) SSA-optimized CNN-LSTM-SEnet wet spraying machine pipe blockage prediction model architecture: This model integrates SSA hyperparameter optimization, CNN spatial feature extraction, LSTM temporal dependency capture and SE attention weighting to achieve efficient processing of multi-feature data.

[0201] (4) End-to-end closed-loop intelligent pipe blockage early warning method for wet spraying machine: A complete closed-loop early warning process was constructed, from pumping parameter acquisition, data cleaning and preprocessing, to training, deployment and online calculation of CNN-LSTM-SEnet model based on SSA optimization, and finally realizing advanced pipe blockage risk prediction and real-time feedback.

[0202] Figure 6 This is a schematic diagram of the structure of the device for constructing the pipe blockage prediction model provided in an embodiment of this application. Figure 6 As shown, the apparatus for constructing the pipe blockage prediction model includes:

[0203] The first acquisition module 601 is used to acquire a first wet spraying machine multi-feature dataset; wherein, the first wet spraying machine multi-feature dataset is used to represent the operating status of the preset wet spraying machine pumping system within a preset first time period, and the end time of the first time period is earlier than the current time.

[0204] The first construction module 602 is used to construct a hybrid model including a feature extraction module, a temporal modeling module, an attention mechanism module, and an output module based on the multi-feature dataset of the first wet spraying machine. The feature extraction module is used to perform local convolution and pooling on the multi-feature dataset of the first wet spraying machine to obtain local spatiotemporal features. The temporal modeling module is used to perform dependency modeling on the local spatiotemporal features to obtain temporal dynamic features. The attention mechanism module is used to evaluate the channel importance and adaptively weight the temporal dynamic features to obtain attention-weighted feature vectors. The output module is used to determine whether pipe blockage or non-blockage occurs in a preset scenario based on the attention-weighted feature vectors.

[0205] The optimization module 603 is used to optimize multiple preset hyperparameters in the hybrid model using a preset search algorithm to obtain a pipe blockage prediction model. The multiple hyperparameters are used to represent the structural configuration and training process of the hybrid model. The search algorithm is used to optimize the multiple hyperparameters so that the prediction performance of the hybrid model meets the preset performance requirements. The pipe blockage prediction model is used to determine whether pipe blockage occurs or not in the scenario.

[0206] In one possible design, the first acquisition module 601 includes:

[0207] The first acquisition unit is used to acquire the multi-feature dataset of the second wet spraying machine.

[0208] The detection unit is used to detect outliers in the multi-feature dataset of the second wet spraying machine and obtain multiple outliers; among them, the multiple outliers are data in the multi-feature dataset of the second wet spraying machine that deviate from the preset data distribution range.

[0209] The removal unit is used to remove multiple outliers from the second wet spraying machine multi-feature dataset to obtain the third wet spraying machine multi-feature dataset.

[0210] The imputation unit is used to impute missing values ​​in the third wet spraying machine multi-feature dataset to obtain the fourth wet spraying machine multi-feature dataset.

[0211] The normalization unit is used to normalize the data in the fourth wet spraying machine multi-feature dataset to obtain the first wet spraying machine multi-feature dataset.

[0212] In one possible design, the device for constructing the pipe blockage prediction model also includes:

[0213] The second acquisition module is used to acquire the multi-feature dataset of the fifth wet spraying machine; wherein, the multi-feature dataset of the fifth wet spraying machine is used to represent the operating status of the wet spraying machine pumping system at the current time.

[0214] The input module is used to input the multi-feature dataset of the fifth wet spraying machine into the pipe blockage prediction model to obtain the pipe blockage prediction result; wherein, the pipe blockage prediction result is used to indicate whether pipe blockage occurs or not in the scenario.

[0215] In one possible design, the input module includes:

[0216] The first input unit is used to input the multi-feature dataset of the fifth wet spraying machine into the feature extraction module to obtain the predicted local spatiotemporal features.

[0217] The second input unit is used to input the predicted local spatiotemporal features into the time series modeling module to obtain the predicted time series dynamic features.

[0218] The third input unit is used to input the predicted temporal dynamic features into the attention mechanism module to obtain the predicted attention weighted feature vector.

[0219] The fourth input unit is used to input the predicted attention-weighted feature vector into the output module to obtain the pipe blockage prediction result.

[0220] In one possible design, the first input unit includes:

[0221] The convolution component is used to perform two-level one-dimensional convolution processing on the multi-feature dataset of the fifth wet spraying machine to obtain preliminary convolution features.

[0222] The pooling component is used to perform two-stage one-dimensional max pooling on the initial convolutional features to obtain the predicted local spatiotemporal features.

[0223] In one possible design, the third input unit includes:

[0224] The average pooling component is used to perform global average pooling on the predicted time-series dynamic features to obtain the channel importance description vector. The predicted time-series dynamic features include multiple feature channels, and the channel importance description vector is used to represent the importance of each feature channel to the pipe blockage prediction.

[0225] The transformation component is used to perform a nonlinear transformation on the channel importance description vector to obtain the original attention weights; where the original attention weights are the preliminary weight values ​​obtained after performing a nonlinear transformation on the channel importance description vector.

[0226] The normalization component is used to normalize the original attention weights to obtain normalized channel attention weights.

[0227] The weighted component is used to multiply the normalized channel attention weights with the predicted temporal dynamic features channel by channel to obtain the attention-weighted feature vector.

[0228] The device for constructing the pipe blockage prediction model provided in this embodiment can execute... Figure 2and Figure 3 The technical solution of the embodiment of the pipe blockage prediction model construction method shown is as follows: its implementation principle and technical effect are similar to... Figure 2 and Figure 3 The method for constructing a pipe blockage prediction model shown is similar to the example provided, and will not be described in detail here.

[0229] Figure 7 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device 70 includes at least one processor 710 and a memory 720. The electronic device also includes a communication component 730. The processor 710, memory 720, and communication component 730 are connected via a bus 740.

[0230] In a specific implementation, at least one processor 710 executes computer execution instructions stored in memory 720, causing at least one processor 710 to implement a method for constructing a pipe blockage prediction model according to the above embodiment.

[0231] The specific implementation process of processor 710 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0232] In the above embodiments, it should be understood that the processor 710 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0233] The memory 720 may include high-speed RAM memory, and may also include non-volatile memory NVM, such as at least one disk storage.

[0234] Bus 740 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Bus 740 can be divided into address bus, data bus, control bus, etc. For ease of illustration, the bus 740 in the accompanying drawings of this application is not limited to only one bus or one type of bus.

[0235] The above description of the functions implemented by electronic devices and main control devices has introduced the solutions provided by the embodiments of the present invention. It is understood that, in order to implement the above functions, the electronic device or main control device includes hardware structures and / or software modules corresponding to the execution of each function. By combining the units and algorithm steps of the various examples described in the embodiments of the present invention, the embodiments of the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the technical solutions of the embodiments of the present invention.

[0236] This application also provides a computer-readable storage medium storing computer-executable instructions. When executed by a processor, these instructions are used to implement a method for constructing a pipe blockage prediction model as described in the above embodiments. In the specific implementation of the aforementioned method for constructing a pipe blockage prediction model, each module can be implemented as a processor.

[0237] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0238] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in application-specific integrated circuits (ASICs). Alternatively, the processor and the readable storage medium can exist as discrete components in an electronic device or a host device.

[0239] This application also provides a computer program product, including a computer program, which, when executed by a processor, is used to implement a method for constructing a pipe blockage prediction model as described in the above embodiments.

[0240] The computer program is stored in a readable storage medium, and at least one processor can read the computer program from the readable storage medium and execute the computer program to perform the scheme provided in any of the above embodiments.

[0241] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.

[0242] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for constructing a plugged tube prediction model, characterized by, The method comprises the following steps: acquiring a first wet sprayer multi-feature data set; wherein the first wet sprayer multi-feature data set is used to represent the running state of a preset wet sprayer pumping system within a preset first time period, and the end time of the first time period is earlier than the current time; constructing a hybrid model comprising a feature extraction module, a time series modeling module, an attention mechanism module, and an output module according to the first wet sprayer multi-feature data set; wherein the feature extraction module is used to perform local convolution and pooling on the first wet sprayer multi-feature data set to obtain local spatio-temporal features, the time series modeling module is used to model the dependency relationship of the local spatio-temporal features to obtain time series dynamic features, the attention mechanism module is used to evaluate the channel importance and adaptively weight the time series dynamic features to obtain an attention weighted feature vector, and the output module is used to determine whether pipe blockage occurs in a preset scene according to the attention weighted feature vector; optimizing a plurality of preset hyperparameters in the hybrid model using a preset search algorithm to obtain a pipe blockage prediction model; wherein the plurality of hyperparameters are used to represent the structure configuration and training process of the hybrid model, the search algorithm is used to optimize the plurality of hyperparameters to make the prediction performance of the hybrid model meet a preset performance requirement, and the pipe blockage prediction model is used to determine whether pipe blockage occurs in the scene.

2. The method of claim 1, wherein, The method further comprises the following steps in the step of acquiring the first wet sprayer multi-feature data set: acquiring a second wet sprayer multi-feature data set; performing outlier detection on the second wet sprayer multi-feature data set to obtain a plurality of outliers; wherein the plurality of outliers are data deviating from a preset data distribution range in the second wet sprayer multi-feature data set; removing the plurality of outliers from the second wet sprayer multi-feature data set to obtain a third wet sprayer multi-feature data set; performing missing value imputation on the third wet sprayer multi-feature data set to obtain a fourth wet sprayer multi-feature data set; normalizing the data in the fourth wet sprayer multi-feature data set to obtain the first wet sprayer multi-feature data set.

3. The method of claim 1, wherein, After the step of optimizing the plurality of preset hyperparameters in the hybrid model using the preset search algorithm to obtain the pipe blockage prediction model, the method further comprises the following steps: acquiring a fifth wet sprayer multi-feature data set; wherein the fifth wet sprayer multi-feature data set is used to represent the running state of the wet sprayer pumping system at the current time; inputting the fifth wet sprayer multi-feature data set into the pipe blockage prediction model to obtain a pipe blockage prediction result; wherein the pipe blockage prediction result is used to represent whether pipe blockage occurs in the scene.

4. The method of claim 3, wherein, The step of inputting the fifth wet sprayer multi-feature data set into the pipe blockage prediction model to obtain the pipe blockage prediction result comprises the following steps: inputting the fifth wet sprayer multi-feature data set into the feature extraction module to obtain predicted local spatio-temporal features; inputting the predicted local spatio-temporal features into the time series modeling module to obtain predicted time series dynamic features; inputting the predicted time series dynamic features into the attention mechanism module to obtain a predicted attention weighted feature vector; Input the predicted attention weighted feature vector into the output module to obtain the pipe blockage prediction result.

5. The method of claim 4, wherein, The fifth wet sprayer multi-feature data set is input into the feature extraction module to obtain a predicted local spatio-temporal feature, including: The fifth wet sprayer multi-feature data set is subjected to two-level one-dimensional convolution processing to obtain preliminary convolution features; The preliminary convolution features are subjected to two-level one-dimensional maximum pooling processing to obtain the predicted local spatio-temporal feature.

6. The method of claim 4, wherein, The predicted time series dynamic feature is input into the attention mechanism module to obtain a predicted attention weighted feature vector, including: The predicted time series dynamic feature is subjected to global average pooling to obtain a channel importance description vector; wherein the predicted time series dynamic feature includes a plurality of feature channels, and the channel importance description vector is used to represent the importance of each feature channel to pipe blockage prediction; The channel importance description vector is subjected to nonlinear transformation to obtain an original attention weight; wherein the original attention weight is a preliminary weight value obtained by nonlinear transformation of the channel importance description vector; The original attention weight is normalized to obtain a normalized channel attention weight; The normalized channel attention weight is multiplied with the predicted time series dynamic feature channel by channel to obtain the attention weighted feature vector.

7. A device for constructing a plugged tube prediction model, characterized by including: The first acquisition module is configured to acquire a first wet sprayer multi-feature data set; wherein the first wet sprayer multi-feature data set is used to represent the running state of a preset wet sprayer pumping system within a preset first time period, and the end time of the first time period is earlier than the current time. The first construction module is configured to construct a hybrid model including a feature extraction module, a time series modeling module, an attention mechanism module, and an output module according to the first wet sprayer multi-feature data set; wherein the feature extraction module is configured to perform local convolution and pooling on the first wet sprayer multi-feature data set to obtain a local spatio-temporal feature, the time series modeling module is configured to model the dependency relationship of the local spatio-temporal feature to obtain a time series dynamic feature, the attention mechanism module is configured to evaluate the channel importance of the time series dynamic feature and adaptively weight it to obtain an attention weighted feature vector, and the output module is configured to determine whether pipe blockage occurs or does not occur in a preset scene according to the attention weighted feature vector. The optimization module is configured to optimize a plurality of preset hyperparameters in the hybrid model using a preset search algorithm to obtain a pipe blockage prediction model; wherein the plurality of hyperparameters are used to represent the structure configuration and training process of the hybrid model, the search algorithm is used to optimize the plurality of hyperparameters to make the prediction performance of the hybrid model meet a preset performance requirement, and the pipe blockage prediction model is used to determine whether pipe blockage occurs or does not occur in the scene.

8. An electronic device, comprising: including: A processor and a memory connected in communication with the processor; The memory stores computer execution instructions; When the processor executes the computer execution instructions stored in the memory, it is used to implement the construction method of the pipe blockage prediction model according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the construction method of the pipe blockage prediction model in any one of claims 1 to 6.

10. A computer program product, characterised in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the construction method of the pipe blockage prediction model in any one of claims 1 to 6.

Images