A leakage identification method and system by identifying liquid form and characteristics

Abstract

The present invention relates to the field of leakage identification technology, and discloses a leakage identification method and system by identifying the morphology and characteristics of liquids, including: obtaining the morphology and characteristic data of the leaking liquid through laser scanning equipment, camera equipment, and sensors; preprocessing the data collected by the three methods respectively, and extracting features from the data after the processing is completed; constructing a multimodal deep learning model for feature fusion, training the model with a training data set, testing after the training is completed, and optimizing and adjusting the model according to the test results; identifying the morphology and characteristics of the leaking liquid, and using an ant algorithm to determine the location of the leakage. The present invention has the following beneficial effects: it is not affected by the type, appearance, and use of the leaking equipment. Unlike traditional leakage identification technology, it is not necessary to select a suitable identification method according to the leaking equipment such as pipes, tanks, valves, walls, etc. By directly identifying the liquid, it is possible to find the leakage point or area.

Description

Technical Field

[0001] The present invention relates to the technical field of leakage identification, and in particular to a leakage identification method and system by identifying the form and characteristics of liquid. Background Art

[0002] At present, some common leakage identification technologies such as infrared imaging, ultrasonic detection, acoustic detection and image recognition-based technologies have their own unique advantages. However, when facing various complex environments, each technology inevitably has its own limitations and is prone to misjudgment and missed judgment. Especially for the power production industry, leakage identification technology often needs to be applied to some high temperature, high pressure, strong electromagnetic interference, strong noise and other environments. At this time, if only one or several leakage identification technologies are relied on, it is difficult to meet the production needs of the power industry, and the accuracy and reliability of the detection results cannot be ensured, and it is impossible to solve the problem of water and oil leakage in pipelines, valves, tanks, and special places in power production. Summary of the invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by the present invention is that the existing leakage identification method cannot meet the needs of the power industry.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a leakage identification method by identifying the form and characteristics of liquid, comprising:

[0006] Obtain the morphology and characteristic data of the leaking liquid through laser scanning equipment, camera equipment, and sensors;

[0007] The data collected by the three methods are preprocessed respectively, and the features of the data are extracted after the processing;

[0008] Build a multimodal deep learning model for feature fusion, use the training data set to train the model, test it after training, and optimize and adjust the model based on the test results;

[0009] Identify the shape and characteristics of the leaking liquid, and use the ant algorithm to determine the location of the leak.

[0010] As a preferred solution of the leakage identification method by identifying the form and characteristics of liquids described in the present invention, wherein: the data collected in the three ways include, when data collection is turned on, automatically performing calibration of the device;

[0011] Acquire the liquid surface shape, contour and position data through laser scanning equipment;

[0012] Acquire image and video data of liquid color, surface shape, and contour through camera monitoring;

[0013] The liquid surface shape, contour, surface fluctuation and conductivity change data are obtained through optical, displacement, resistance and temperature sensors.

[0014] As a preferred solution of the leakage identification method by identifying the liquid form and characteristics of the present invention, wherein: the preprocessing includes preprocessing the data collected in three ways respectively;

[0015] Preprocessing of data acquired by laser scanning equipment through filtering, noise reduction, surface data smoothing, and liquid contour recognition;

[0016] Preprocess the image data acquired by the camera device through noise reduction, correction, contrast enhancement, and image segmentation; Preprocess the video data acquired by the camera device through noise reduction, correction, and video frame cutting;

[0017] The data obtained by the sensor is preprocessed through denoising, data cleaning, smoothing and transformation.

[0018] As a preferred solution of the leakage identification method by identifying the liquid morphology and characteristics described in the present invention, wherein: the feature extraction includes, the feature extraction of the laser scanning device data includes, using the data obtained by the laser scanning device to construct a three-dimensional point cloud model of the liquid surface; specifically:

[0019] Analyze the geometric features of the 3D point cloud model and calculate the surface roughness index of the point cloud. The formula is expressed as:

[0020]

[0021] in, Indicates the point cloud The height value of each point; Indicates the average value of the point cloud height; Indicates the total number of points in the point cloud;

[0022] 3D point cloud surface curvature including Gaussian curvature and mean curvature , the formula is:

[0023]

[0024]

[0025] in, Represents the principal curvature of a surface point in a direction; Indicates that the surface point is perpendicular The principal curvature of the direction; Represents the slope of the surface point in the x direction; Represents the slope of the surface point in the y direction;

[0026] The volume and surface area formula is expressed as:

[0027]

[0028] in, Represents a triangular mesh Three vertices of; each triangle is composed of adjacent point cloud points;

[0029] Estimate the overall surface area by summing the areas of all triangles, surface area The formula is expressed as:

[0030]

[0031] Feature extraction of camera equipment data includes extracting color features, texture features, and shape contour features of the liquid;

[0032] Extract the color histogram of the leaking liquid and calculate the color moment. The formula is:

[0033]

[0034] in, Indicates color counts in histograms; Indicates the total number of pixels; Indicates the average value of the color; Indicates the standard deviation of color; gray-level co-occurrence matrix is ​​used to analyze the texture characteristics of the liquid surface, and the local binary pattern method is used to extract the surface texture information;

[0035] Statistical analysis methods are used to extract sensor device data features.

[0036] As a preferred solution of the leakage identification method by identifying the liquid morphology and characteristics described in the present invention, wherein: the construction of the multimodal deep learning model includes using a multilayer perceptron to learn the relationship between different features and perform feature fusion. The multilayer perceptron structure is represented as:

[0037] The input layer has different data sources, and the input feature vector of each data source is ;

[0038] Hidden layers, the first hidden layer applies a transformation to each input feature vector, and the output of the first hidden layer , expressed as:

[0039]

[0040] in, represents the weight of the first hidden layer; represents the bias of the first hidden layer, represents the activation function;

[0041] The feature fusion layer fuses the first hidden layer outputs of all data sources to form a comprehensive feature vector , the formula is:

[0042]

[0043] Apply further transformation to the fused feature vector to generate comprehensive features, and the fusion layer outputs , the formula is:

[0044]

[0045] in, Represents the weight of the fusion layer; represents the bias of the fusion layer;

[0046] The output layer performs linear output of classification and regression tasks according to specific application tasks. The formula is expressed as:

[0047]

[0048] in, represents the activation function of the output layer; represents the weights and Represents the bias of the output layer.

[0049] As a preferred solution of the leakage identification method by identifying the liquid morphology and characteristics of the present invention, wherein: the optimization and adjustment of the model includes dividing the data set, randomly dividing the fused data set into 70% training set, 15% validation set and 15% test set;

[0050] Use the early stopping method, set monitoring indicators, including accuracy and loss rate, train the model using the training set without applying early stopping, record the number of cycles and final performance of each training, and average them to get the performance baseline;

[0051] Based on the performance baseline, set the initial patience threshold and evaluate the model’s performance on the validation set at the end of each epoch.

[0052] If the model performance does not improve, the counter is incremented; if the performance improves, the counter is reset;

[0053] Determine whether to stop early. If the model performance improvement for multiple consecutive cycles is not greater than the performance baseline, and the counter count reaches the set patience threshold, stop training;

[0054] Evaluate model performance on an independent test set, monitor and prevent overfitting, ensure model generalization, and deploy trained models to production monitoring environments for real-time identification of liquid leaks.

[0055] As a preferred solution of the leakage identification method by identifying the morphology and characteristics of liquids described in the present invention, wherein: the determination of the type and leakage location of the leaking liquid includes using a multimodal deep learning model to identify the morphological characteristics of the leaking liquid and using an ant algorithm to identify the leaking liquid;

[0056] Define the search space, analyze the possible types of leaking liquid, and divide the possible paths of the leaking liquid types into multiple nodes; define the potential types of the leaking liquid as connecting paths and leaking nodes;

[0057] Initialize pheromones, pheromones on each path , initialized to a constant value ;

[0058]

[0059] in, Indicates the path from the node To Node The pheromone concentration is the distribution probability of the leakage liquid type according to the identified leakage liquid morphology and characteristics; the algorithm parameters are set, and the ant colony size , pheromone volatility , Factors affecting pheromone intensity and heuristic impact factor ;

[0060] Path construction and search include,simulating ant behavior, each ant selects the next node based on the current pheromone concentration and heuristic information;,selection probability The formula is:

[0061]

[0062] in, Represents heuristic information; Represents ants An optional set of next nodes; Represents ants optional next node;

[0063] Pheromone update, local update rule, whenever the ant passes through the path arrive After that, the pheromone is updated to simulate the natural volatilization of pheromones and the influence of the frequency of path use. The update formula is expressed as:

[0064]

[0065] in, represents the increment of pheromone left by ants on the path, which is proportional to the quality of the path;

[0066] The global update rule includes updating the pheromone based on the search results of all ants after each round of iteration and strengthening the pheromone on the optimal path. The formula is expressed as:

[0067]

[0068] in, Represents the pheromone increment on the optimal path found in this round of iteration;

[0069] When two or more collection devices detect leaking liquid, the path given to the current area , speed up the identification of leaking liquid; perform repeated iterations, repeatedly execute the path construction and pheromone update process, and use the path with the highest pheromone concentration to determine the most likely category of leaking liquid;

[0070] After identifying the category of leaking liquid, the category of leaking liquid is replaced by the leakage point, and the pheromone concentration is the distribution probability of the leakage location; the collection devices are paths, and the collection devices are nodes, and the ant algorithm is used to search for the leakage location; repeated iterations are performed, and the path construction and pheromone update processes are repeatedly executed, and the most likely leakage location is determined using the path with the highest pheromone concentration.

[0071] A leakage identification system by identifying the form and characteristics of liquid using any method as described in the present invention, wherein:

[0072] Data acquisition module, sensor integration, integrates multiple sensors for real-time monitoring and data collection; receives raw data from various sensors and transmits it to the data processing and feature extraction module;

[0073] The data processing and feature extraction module pre-processes the collected data, performs filtering, noise reduction, correction and enhancement processing; extracts key features from the processed data, including the surface tension, roughness, curvature, color, texture, surface area, fluctuation, density, viscosity, temperature and conductivity of the liquid;

[0074] Model calculation module, using deep learning model, builds and trains multimodal deep learning model to identify and classify the characteristics and state of liquid; integrates ant colony algorithm to optimize the feature selection and parameters of the model, and uses ant colony algorithm to predict the flow path of liquid in pipes or containers and determine the leakage point;

[0075] The decision-making and response module determines whether leakage occurs and the specific location of the leakage point based on the output of the deep learning model and the ant colony algorithm, and triggers corresponding early warning and response measures according to the judgment results.

[0076] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein: the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0077] A computer-readable storage medium stores a computer program, comprising: when the computer program is executed by a processor, the steps of implementing any one of the methods of the present invention are implemented.

[0078] The beneficial effects of the present invention are as follows: it has little limitation in use and is not affected by the type, appearance and purpose of the leakage equipment. Unlike the traditional leakage identification technology, it is not necessary to select a suitable identification method according to the leakage equipment such as pipes, tanks, valves, walls, etc. By directly identifying the liquid, it is possible to find the leakage point or area. At the same time, in order to accurately identify the shape and characteristics of the liquid, this patent adopts a combination of multiple identification technologies and machine learning. Compared with a single leakage identification technology, the identification accuracy and reliability are higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0080] Figure 1 An overall flow chart of a leakage identification method by identifying the liquid morphology and characteristics provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0081] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0082] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a leakage identification method by identifying the form and characteristics of liquid, comprising:

[0083] S1: Obtain the morphology and characteristic data of the leaking liquid through laser scanning equipment, camera equipment, and sensors.

[0084] Furthermore, temperature and humidity sensors are deployed to monitor environmental conditions in real time. The system automatically adjusts scanning parameters based on the data from these sensors. Standard sample scanning: Calibration is performed using standard samples with known size and reflective properties. The samples should be selected to cover the optical properties within the operating range.

[0085] Furthermore, calibration is automatically performed during the acquisition step, and the calibration coefficients Based on standard values and the original measurement The ratio is used to adjust the measured value to the standard value range, and the formula is expressed as:

[0086]

[0087] Furthermore, the correction function of the environment and liquid properties is the environmental correction and liquid property correction It is calculated based on real-time monitored environmental parameters and known properties of the liquid, and the formula is expressed as:

[0088]

[0089]

[0090] in, and Respectively represent temperature and humidity; and Indicates the density and viscosity of liquids; ,and represents the experimentally determined coefficients. The comprehensive calibration formula combines these into a comprehensive calibration formula that adjusts the device reading to more accurately reflect the actual properties of the liquid:

[0091]

[0092] Furthermore, the deviation between the actual scan data and the standard data is calculated. The laser intensity and receiver sensitivity are adjusted until the measured data is consistent with the standard sample data. Calibration verification: Repeat the above steps until the error of the system's measurement results for several consecutive times is within an acceptable range.

[0093] Furthermore, the data collected by the three methods include: equipment calibration before collecting data; the laser scanning device uses a multi-wavelength laser to scan the leaking liquid, and uses data fusion on the various laser data generated by the scanning to obtain a laser scanning device data set.

[0094] Furthermore, the liquid surface shape, contour and position data are acquired through laser scanning equipment; the image and video data of the liquid color, surface shape and contour are acquired through camera equipment; the liquid surface shape, contour, surface fluctuation and conductivity change data are acquired through optical, displacement, resistance and temperature sensors.

[0095] It should be noted that the multi-wavelength laser system can improve the adaptability to liquids with different optical properties by using multi-wavelength lasers. Wavelength selection: Select different laser wavelengths, each wavelength corresponds to different material absorption and reflection characteristics. Synchronous scanning: All wavelengths of lasers scan the same liquid surface at the same time to collect data on reflected or transmitted light.

[0096] S2: Preprocess the data collected by the three methods respectively, and perform feature extraction and feature fusion on the data after processing.

[0097] Furthermore, the preprocessing includes preprocessing the data collected by the laser scanning device using filtering, noise reduction, smoothing surface data and identifying liquid contours.

[0098] Furthermore, for the data collected by the camera device, the image data is preprocessed using noise reduction, correction, contrast enhancement, and image segmentation; the video data is preprocessed using noise removal, correction, and video frame cutting.

[0099] Furthermore, the data collected by optical, displacement, resistance and temperature sensors are pre-processed using denoising, data cleaning, smoothing and transformation.

[0100] Furthermore, the image segmentation includes applying graph cut technology to separate the liquid part from the background; the transformation includes performing mathematical transformation on the sensor data to linearize the non-linear relationship.

[0101] Furthermore, the feature extraction includes: the feature extraction of the laser scanning device data includes: constructing a three-dimensional point cloud model of the liquid surface using the data obtained by the laser scanning device; analyzing the geometric features of the three-dimensional point cloud model, and calculating the surface roughness index of the point cloud, which is expressed as follows:

[0102]

[0103] in, Indicates the point cloud The height value of each point; Indicates the average value of the point cloud height; Represents the total number of points in the point cloud.

[0104] Furthermore, the surface curvature of the 3D point cloud includes Gaussian curvature and mean curvature , the formula is:

[0105]

[0106]

[0107] in, and Represents the principal curvature of the surface at a point.

[0108] Going further, the volume and surface area formulas are expressed as:

[0109]

[0110] in, Represents a triangular mesh The three vertices of each triangle are composed of adjacent point cloud points, and the overall surface area is estimated by calculating the sum of the areas of all triangles. The formula is expressed as:

[0111]

[0112] Furthermore, the feature extraction of the camera data includes extracting the color features, texture features, and shape contour features of the liquid; extracting the color histogram of the leaking liquid, and calculating the color moment. The first-order moment (mean) formula is expressed as:

[0113]

[0114] in: is the color value; the second-order moment (variance) formula is expressed as:

[0115]

[0116] The third order moment formula is expressed as;

[0117] in, Indicates the frequency or probability of occurrence of color values; Indicates color counts in histograms; Indicates the total number of pixels; Indicates the average value of the color; Represents the standard deviation of color; the gray level co-occurrence matrix is ​​used to analyze the texture characteristics of the liquid surface, and the local binary pattern method is used to extract the surface texture information.

[0118] Furthermore, feature extraction of sensor device data includes applying statistical analysis to extract features.

[0119] It should be noted that mean curvature is used to characterize the average curvature of a surface near a point. It is particularly important in surface modeling, graphic design, and material science because mean curvature is directly related to the surface tension of an object. For example, the shape of a droplet is affected by internal and external pressures, which in turn are related to mean curvature.

[0120] S3: Build a multimodal deep learning model, train the model using a training dataset, test the model after training, and optimize and adjust the model based on the test results.

[0121] Furthermore, the construction of the multimodal deep learning model includes feature fusion, including standardized features, unified feature scales; excluding redundant or irrelevant features, and screening through a feature importance scoring algorithm based on a decision tree.

[0122] Furthermore, a multi-layer perceptron is used to learn the relationship between different features, perform feature fusion, and construct a multi-layer perceptron structure.

[0123] Furthermore, the input layer has different data sources, and the input feature vector of each data source is .

[0124] Furthermore, the hidden layer, the first hidden layer applies a transformation to each input feature vector, and the output of the first hidden layer , expressed as:

[0125]

[0126] in, represents the weight of the first hidden layer; represents the bias of the first hidden layer, Represents the activation function.

[0127] Furthermore, the feature fusion layer fuses the first hidden layer outputs of all data sources to form a comprehensive feature vector , the formula is:

[0128]

[0129] Furthermore, further transformation is applied to the fused feature vector to produce comprehensive features, and the fusion layer outputs , the formula is:

[0130]

[0131] in, Represents the weight of the fusion layer; Represents the bias of the fusion layer.

[0132] Furthermore, the output layer performs linear output of classification and regression tasks according to specific application tasks, and the formula is expressed as:

[0133]

[0134] in, represents the activation function of the output layer; represents the weights and Represents the bias of the output layer.

[0135] Furthermore, the optimization and adjustment of the model includes data set division, dividing the fused data set into 70% training set, 15% validation set and 15% test set.

[0136] Furthermore, we use early stopping techniques, set monitoring indicators, including accuracy and loss rate, train the model using the training set without applying early stopping, record the number of cycles and final performance of each training, and average them to get the performance baseline.

[0137] Furthermore, we set an initial patience threshold based on the performance baseline, and evaluate the model’s performance on the validation set at the end of each epoch. If the model performance does not improve, we increment the counter; if the performance improves, we reset the counter.

[0138] Furthermore, it is determined whether to stop early. If the model performance improvement for multiple consecutive cycles is not greater than the performance baseline, and the counter count reaches the set patience threshold, the training is stopped.

[0139] Furthermore, the model performance is evaluated on an independent test set to monitor and prevent overfitting, ensure the generalization ability of the model, and deploy the trained model to the production monitoring environment for real-time identification of liquid leaks.

[0140] It should be noted that early stopping not only helps prevent overfitting, but also ensures that the model is not trained too much, which can save resources and reduce training time. By monitoring the performance (such as loss or accuracy) on the validation set, you can more objectively evaluate the generalization ability of the model to unseen data. Using early stopping can also avoid wasting resources during training, especially when using expensive computing resources.

[0141] S4: Identify the shape and characteristics of the leaking liquid, and use the ant algorithm to determine the point or area where the leakage occurs.

[0142] Furthermore, the determining the type and leakage location of the leaking liquid includes using a multimodal deep learning model to identify the morphological features of the leaking liquid and using an ant algorithm to identify the leaking liquid;

[0143] Define the search space, analyze the possible types of leaking liquid, and divide the possible paths of the leaking liquid types into multiple nodes; define the potential types of the leaking liquid as connecting paths and leaking nodes;

[0144] Initialize pheromones, pheromones on each path , initialized to a constant value ;

[0145]

[0146] in, Indicates the path from the node To Node The pheromone concentration is the distribution probability of the leakage liquid type according to the identified leakage liquid morphology and characteristics; the algorithm parameters are set, and the ant colony size , pheromone volatility , Factors affecting pheromone intensity and heuristic impact factor ;

[0147] Path construction and search include,simulating ant behavior, each ant selects the next node based on the current pheromone concentration and heuristic information;,selection probability The formula is:

[0148] ,

[0149] in, Represents heuristic information; Represents ants An optional set of next nodes; Represents ants optional next node;

[0150] Pheromone update, local update rule, whenever the ant passes through the path arrive After that, the pheromone is updated to simulate the natural volatilization of pheromones and the influence of the frequency of path use. The update formula is expressed as:

[0151] ,

[0152] in, represents the increment of pheromone left by ants on the path, which is proportional to the quality of the path;

[0153] The global update rule includes updating the pheromone based on the search results of all ants after each round of iteration and strengthening the pheromone on the optimal path. The formula is expressed as:

[0154]

[0155] in, Represents the pheromone increment on the optimal path found in this round of iteration;

[0156] When two or more collection devices detect leaking liquid, the path given to the current area , speed up the identification of leaking liquid; perform repeated iterations, repeatedly execute the path construction and pheromone update process, and use the path with the highest pheromone concentration to determine the most likely category of leaking liquid;

[0157] After identifying the category of leaking liquid, the category of leaking liquid is replaced by the leakage point, and the pheromone concentration is the distribution probability of the leakage location; the collection devices are paths, and the collection devices are nodes, and the ant algorithm is used to search for the leakage location; repeated iterations are performed, and the path construction and pheromone update processes are repeatedly executed, and the most likely leakage location is determined using the path with the highest pheromone concentration.

[0158] It should be noted that the integration of deep learning models and ant colony algorithms requires not only technical integration, but also ensuring that data flows and information are seamlessly connected between the two. Deep learning models are responsible for extracting key features and identifying patterns from multimodal data, while ant colony algorithms use this information to optimize the search and decision-making process. Therefore, ensuring effective information exchange between the two is key.

[0159] On the other hand, the present embodiment further provides a leakage identification system for identifying the form and characteristics of a liquid, which includes:

[0160] Data acquisition module, sensor integration, integrates multiple sensors for real-time monitoring and data collection; receives raw data from various sensors and transmits it to the data processing and feature extraction module.

[0161] The data processing and feature extraction module pre-processes the collected data, filters, reduces noise, corrects and enhances it; and extracts key features from the processed data, including the surface tension, roughness, curvature, color, texture, surface area, fluctuation, density, viscosity, temperature and conductivity of the liquid.

[0162] The model calculation module uses a deep learning model to build and train a multimodal deep learning model to identify and classify the characteristics and states of the liquid; integrates the ant colony algorithm to optimize the feature selection and parameters of the model, and uses the ant colony algorithm to predict the flow path of the liquid in the pipe or container and determine the leakage point.

[0163] The decision-making and response module determines whether leakage occurs and the specific location of the leakage point based on the output of the deep learning model and the ant colony algorithm, and triggers corresponding early warning and response measures according to the judgment results.

[0164] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0165] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0166] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0167] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0168] Example 2: The following is an embodiment of the present invention, which provides a leakage identification method by identifying the liquid morphology and characteristics. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0169] In this example, the goal is to verify the effectiveness of the proposed multimodal deep learning model combined with the ant algorithm for identifying and locating liquid leakage. First, three different data acquisition devices are set up: laser scanning equipment, camera equipment, and various physical sensors to collect the morphology and characteristic data of the liquid.

[0170] Laser scanning equipment is used to obtain 3D shape and contour data of liquid surfaces. The equipment provides detailed images of the roughness and morphology of the liquid surface by accurately measuring tiny height changes on the liquid surface.

[0171] The camera equipment records the color changes and flow state of the liquid. Through high-speed camera technology, continuous frames of liquid movement can be captured to further analyze the dynamic characteristics of the liquid.

[0172] Sensors include temperature, pressure and conductivity sensors, which monitor changes in the physical and chemical properties of the liquid and provide real-time data on the liquid's state.

[0173] The data preprocessing stage includes denoising and smoothing of laser scanning data, frame rate adjustment and contrast enhancement of video data, and standardization and outlier processing of sensor data. After that, various data are input into the feature extraction module to extract key parameters such as liquid surface tension, color change, flow rate, temperature change, etc.

[0174] Next, a multimodal deep learning model was built that fuses data from different sensors and processes and analyzes this information through a unified framework. The model was first trained on 70% of the data, 15% of the data was used to verify the model performance, and the remaining 15% of the data was used as a final test. Early stopping techniques were used to avoid overfitting, and monitoring indicators included accuracy and loss rate.

[0175] Finally, the ant algorithm is used to optimize the model output path and accurately predict the leakage location. The ant algorithm simulates the behavior of ants looking for food sources based on the feature strength of the multimodal data output and determines the path with the highest pheromone concentration, that is, the most likely leakage point. The algorithm performance experiment is shown in Table 1, and the results of the simulation technology comparison experiment are shown in Table 2.

[0176] Table 1 Algorithm performance comparison table

[0177] Algorithm Type Recognition accuracy (%) Positioning time (ms) System computing capacity (GFLOPS) Error rate (%) Machine Learning Models 90 200 50 8 Sound wave recognition algorithm 87 1000 70 10 Support Vector Machine Algorithm 85 350 30 12 Decision Tree Algorithm 80 50 100 15 Multimodal Deep Learning Models 95 10 20 5

[0178] Table 2 Comparison of experimental results

[0179] Technology Type Recognition accuracy (%) Positioning time (ms) Maintenance frequency (times / year) Energy consumption (kWh / year) Automation level (%) Traditional video surveillance technology 78 60 12 500 20 Pressure sensing based monitoring 82 50 12 1500 40 Intelligent recognition system 88 40 6 2200 65 Multimodal integrated recognition system 95 12 3 1200 85

[0180] The multimodal deep learning + ant algorithm invented by us has achieved 95% recognition accuracy and a positioning time of only 12 seconds, which is far superior to other single algorithms and existing technologies. For example, the recognition accuracy of the support vector machine (SVM) is 85% and the positioning time is 30 seconds, while the positioning time of the pressure sensor-based monitoring system is 50 seconds. This shows that the method combining deep learning and ant algorithms makes more effective use of data features and improves processing speed and accuracy.

[0181] The maintenance frequency of the multimodal integrated recognition system is 3 times per year, which is lower than the 6 times of the multimodal system, showing lower maintenance requirements and longer operation stability. At the same time, the energy consumption is 1200 kWh / year, which is significantly lower than the 2200 kWh / year of the existing multimodal system. The lower maintenance frequency and energy consumption not only reduce operating costs, but also reduce the impact on the environment.

[0182] The error rate of the multimodal comprehensive recognition system is only 5%, which is much lower than the 15% of the decision tree algorithm and the 20% of the traditional video surveillance technology. In addition, the automation level is as high as 85%, which reduces the need for human operation, reduces the possibility of errors, and improves operational efficiency.

[0183] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A leakage identification method by identifying the form and characteristics of a liquid, characterized in that: include: Obtain the morphology and characteristic data of the leaking liquid through laser scanning equipment, camera equipment, and sensors; The data collected by the three methods are preprocessed respectively, and the features of the data are extracted after the processing; Build a multimodal deep learning model for feature fusion, use the training data set to train the model, test it after training, and optimize and adjust the model based on the test results; Identify the shape and characteristics of the leaking liquid, and use the ant colony algorithm to determine the type and location of the leaking liquid; The feature extraction includes: the feature extraction of laser scanning device data includes: using the data acquired by the laser scanning device to construct a three-dimensional point cloud model of the liquid surface; specifically: Analyze the geometric features of the 3D point cloud model and calculate the surface roughness index of the point cloud. The formula is expressed as: Among them, z i Represents the height value of the i-th point in the point cloud; represents the average value of the point cloud height; n represents the total number of points in the point cloud; The surface curvature of the three-dimensional point cloud includes Gaussian curvature K and mean curvature H, and the formula is expressed as: Where k1 represents the principal curvature of the surface point in one direction; k2 represents the principal curvature of the surface point in the direction perpendicular to k1; z x represents the slope of the surface point in the x direction; z y Represents the slope of the surface point in the y direction; The volume formula is: Among them, p i ,q i ,r i Represents the three vertices of the triangular mesh i; each triangle is composed of adjacent point cloud points; The overall surface area is estimated by calculating the sum of the areas of all triangles. The formula for surface area A is: Feature extraction of camera equipment data includes extracting color features, texture features, and shape contour features of the liquid; Extract the color histogram of the leaking liquid and calculate the color moment. The formula is: Among them, H s (c i ) represents color c i The count in the histogram; N represents the total number of pixels; μ represents the average value of the color; σ represents the standard deviation of the color; the gray level co-occurrence matrix is ​​used to analyze the texture characteristics of the liquid surface, and the local binary pattern method is used to extract the surface texture information; Use statistical analysis methods to extract sensor device data features; The construction of the multimodal deep learning model includes using a multilayer perceptron to learn the relationship between different features and perform feature fusion. The structure of the multilayer perceptron is represented as follows: The input layer has m different data sources, and the input feature vector of each data source is x i ; Hidden layers, the first hidden layer applies a transformation to each input feature vector, and the output of the first hidden layer It is expressed as: in, represents the weight of the first hidden layer; represents the bias of the first hidden layer, and f represents the activation function; The feature fusion layer fuses the first hidden layer outputs of all data sources to form a fused comprehensive feature vector h (fused) , the formula is: Further transformation is applied to the fused feature vector to generate comprehensive features, and the fusion layer outputs h (2) , the formula is: h (2) =f(W (2) h (fused) +b (2) ) Among them, W (2) represents the weight of the fusion layer; b (2) represents the bias of the fusion layer; The output layer performs linear output of classification and regression tasks according to specific application tasks. The formula is expressed as: y=g(W (out) h (2) +b (out) ) Among them, g represents the activation function of the output layer; W (out) represents the weights of the output layer and b (out) represents the bias of the output layer; Determining the type and leakage location of the leaking liquid includes using a multimodal deep learning model to identify the morphological characteristics of the leaking liquid and using an ant colony algorithm to identify the leaking liquid; Define the search space, analyze the possible types of leaking liquid, and divide the possible paths of the leaking liquid types into multiple nodes; define the potential types of the leaking liquid as connecting paths and leaking nodes; Initialize pheromone, pheromone τ on each path ij , initialized to a constant value τ0; t ij (0)=τ0 Among them, τ ij represents the pheromone concentration of the path from node i to node j, that is, the distribution probability of the leakage liquid type is determined according to the identified leakage liquid morphology and characteristics; sets the algorithm parameters, ant colony size n0, pheromone volatility rate ρ, pheromone intensity influence factor α and heuristic influence factor β; Path construction and search include,simulating ant behavior, each ant selects the next node based on the current pheromone concentration and heuristic information;,selection probability The formula is: Among them, η ij Indicates heuristic information; allowed k represents the set of next nodes that ant k can choose; l represents the next node that ant k can choose; Pheromone update, local update rule, every time the ant passes through path i to j, the pheromone is updated to simulate the natural volatilization of pheromones and the influence of path usage frequency. The update formula is expressed as: t ij =(1-ρ)·τ ij +ρ·Δτ ij Among them, Δτ ij represents the increment of pheromone left by ants on the path, which is proportional to the quality of the path; The global update rule includes updating the pheromone based on the search results of all ants after each round of iteration and strengthening the pheromone on the optimal path. The formula is expressed as: in, Represents the pheromone increment on the optimal path found in this round of iteration; When two or more collection devices detect leaking liquid, the path given to the current area Accelerate the identification of leaking liquids; perform repeated iterations, repeatedly execute the path construction and pheromone update process, and use the path with the highest pheromone concentration to determine the most likely leaking liquid category; After identifying the category of the leaking liquid, the leaking liquid category is replaced by the leakage point, and the pheromone concentration is the distribution probability of the leakage location; the collection devices are paths, and the collection devices are nodes. The ant colony algorithm is used to search for the leakage location; repeated iterations are performed, and the path construction and pheromone update processes are repeatedly executed, and the most likely leakage location is determined using the path with the highest pheromone concentration.

2. The method for identifying leakage by identifying the form and characteristics of liquid according to claim 1, characterized in that: The data collected in the three ways include automatically performing calibration of the device when data collection is turned on; Acquire the liquid surface shape, contour and position data through laser scanning equipment; Acquire image and video data of liquid color, surface shape, and contour through camera monitoring; The liquid surface shape, contour, surface fluctuation and conductivity change data are obtained through optical, displacement, resistance and temperature sensors.

3. The method for identifying leakage by identifying the form and characteristics of liquid according to claim 2, characterized in that: The preprocessing includes preprocessing the data collected in three ways respectively; Preprocessing of data acquired by laser scanning equipment through filtering, noise reduction, surface data smoothing, and liquid contour recognition; Preprocess the image data acquired by the camera device through noise reduction, correction, contrast enhancement, and image segmentation; Pre-process the video data acquired by the camera device through denoising, correction, and video frame cutting; The data obtained by the sensor is preprocessed through denoising, data cleaning, smoothing and transformation.

4. The method for identifying leakage by identifying the form and characteristics of liquid according to claim 3, characterized in that: The optimization and adjustment of the model includes dividing the data set, randomly dividing the fused data set into 70% training set, 15% validation set and 15% test set; Use the early stopping method, set monitoring indicators, including accuracy and loss rate, train the model using the training set without applying early stopping, record the number of cycles and final performance of each training, and average them to get the performance baseline; Based on the performance baseline, set the initial patience threshold and evaluate the model’s performance on the validation set at the end of each epoch. If the model performance does not improve, the counter is incremented; if the performance improves, the counter is reset; Determine whether to stop early. If the model performance improvement for multiple consecutive cycles is not greater than the performance baseline, and the counter count reaches the set patience threshold, stop training; Evaluate model performance on an independent test set, monitor and prevent overfitting, ensure model generalization, and deploy trained models to production monitoring environments for real-time identification of liquid leaks.

5. A leakage identification system by identifying the form and characteristics of liquid using the method according to any one of claims 1 to 4, characterized in that: Data acquisition module, sensor integration, integrates multiple sensors for real-time monitoring and data collection; receives raw data from various sensors and transmits it to the data processing and feature extraction module; The data processing and feature extraction module pre-processes the collected data, including filtering, noise reduction, correction and enhancement; Extract key features from processed data, including the liquid’s surface tension, roughness, curvature, color, texture, surface area, undulation, density, viscosity, temperature, and conductivity; Model calculation module, using deep learning model, builds and trains multimodal deep learning model to identify and classify the characteristics and state of liquid; integrates ant colony algorithm to optimize the feature selection and parameters of the model, and uses ant colony algorithm to predict the flow path of liquid in pipes or containers and determine the leakage point; The decision-making and response module determines whether leakage occurs and the specific location of the leakage point based on the output of the deep learning model and the ant colony algorithm, and triggers corresponding early warning and response measures according to the judgment results.

6. A computer device comprising: Memory and processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the leakage identification method by identifying the liquid form and characteristics as described in any one of claims 1-4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the leakage identification method by identifying the liquid morphology and characteristics as described in any one of claims 1 to 4 are implemented.

Images