Compression model and dictionary model method for detecting pipeline water leakage signal
Through deep learning, multi-layer single dictionary model is constructed, and layer-by-layer training and encoding is solved, and the calculation complexity and parameter adjustment difficulty of K-SVD dictionary when detecting leaky signals in pipelines is solved, achieving more efficient signal reconstruction and generalization capabilities , suitable for actual water leakage monitoring.
Patent Information
- Application Number
- CN202411345038.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2025-01-10
AI Technical Summary
The existing K-SVD dictionary has high computational complexity when detecting pipeline leakage signals, is difficult to adjust parameters, and is prone to falling into local optimal solutions. The observation matrix design needs to meet the finite isometric properties, which limits the effectiveness of signal reconstruction.
Using a multi-layer single dictionary model, deep learning layer by layer training, a deep dictionary model is built, and multiple dictionary layers are used to encode and decode signals layer by layer, reduce the computational complexity and improve generalization ability, and ensure convergence using greedy iterative learning method. .
It improves the reconstruction accuracy and generalization of leaked signals, reduces the operation difficulty, and is suitable for compression and reconstruction tasks in actual leak monitoring.
Smart Images
Figure CN119298915A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of pipeline leakage detection, in particular to a compression model and a dictionary model method for detecting pipeline leakage signals. Background Art
[0002] At present, with the increasing shortage of fresh water resources and the continuous expansion of water supply networks, pipeline leakage has become one of the urgent problems to be solved in water resource management. With the advancement of science and technology, the widespread application of Internet of Things (IoT) technology and Compressed Sensing (CS) theory has provided new ideas for solving pipeline leakage. In the IoT system for detecting leakage, due to the limited storage and processing capabilities of the nodes, the Nyquist sampling theorem faces the limitation of high-frequency sampling applicability. The compressed sensing theory is applied to the terminal nodes for compressed sampling, and the aggregation nodes are used to reconstruct the signal.
[0003] The most critical technologies in compressed sensing theory mainly include the sparse representation of signals, the design of observation matrices and reconstruction algorithms. Due to the complexity of the actual environment for detecting water leakage, the sparse representation of signals uses the K-SVD dictionary instead of the fixed sparse basis. Although it shows its advantage of strong generalization ability in application, it also has shortcomings. Specifically, the computational complexity of reconstructing signals using the K-SVD dictionary is high; K-SVD is sensitive to the initial dictionary, number of iterations, regularization parameters, etc. The setting of these parameters requires fine adjustment, which increases the difficulty of operation in practical applications; the K-SVD algorithm is prone to fall into the problem of local optimal solutions, which limits its ability to find the global optimal solution. In addition, the design of the observation matrix should meet the conditions of the finite isometry property (RIP) to effectively reconstruct the signal. Summary of the invention
[0004] Therefore, in order to solve the above-mentioned shortcomings, the present invention provides a compression model and a dictionary model method for detecting pipeline leakage signals.
[0005] The present invention is implemented by constructing a compression model and a dictionary model method for detecting pipeline water leakage signals, comprising the following steps: Collect diverse data and conduct multi-layer single dictionary model learning on the computer to obtain compression model and dictionary model; Transplant the compression model at the terminal node to compress and sample the collected signal; The dictionary model is transplanted at the sink node to reconstruct the received compressed signal.
[0006] Preferably, multiple dictionary layers constitute the architecture of the deep dictionary model, and each dictionary learning layer consists of a dictionary and a nonlinear activation function; The input data is first encoded by the first dictionary learning layer, then passed to the next dictionary learning layer, and so on until it reaches the last dictionary learning layer; The output of the last dictionary learning layer is a sparse representation of the data; The layer-by-layer training process of deep dictionary learning is as follows: First, train and learn the feature representation Z1 and dictionary D1 of the first layer: X=D1×Z1; Then, the feature Z1 of the first layer is decomposed to solve the dictionary D2 and feature Z2 of the second layer: Z1=D2×Z2.
[0007] Preferably, multiple single dictionaries are connected together through a deep learning framework to form a multi-layer single dictionary model; The synthesis framework of single dictionary learning enables signal features to synthesize data along the direction of the dictionary. The principle is shown in the formula; X=D1Z Where X represents the input data, D1 represents the dictionary to be learned, and Z is the sparse coefficient of X on the dictionary D1; The single dictionary learning model is extended to a two-layer dictionary learning model, as shown in the formula; X=D1D2Z2 Where X represents the input data, D1 represents the first-layer dictionary to be learned, D2 represents the second-layer dictionary to be learned, the lines of the same color represent atoms in the same column of the current layer dictionary, and Z2 is the sparse coefficient of X on the second-layer dictionary; The dictionary is obtained by adopting a greedy iterative learning method, and each layer is learned separately to ensure convergence until the third layer dictionary training is completed.
[0008] Preferably, the layer-by-layer training of the deep dictionary model includes the following steps: First, initialize the parameters of the first dictionary layer, including the dictionary and nonlinear activation function; Then, training is performed layer by layer, with each layer encoding the input data into a sparse representation through forward propagation and updating the parameters through backpropagation; After each layer is trained, the learned features are used as the input of the next layer, and the above steps are repeated until the required number of layers or convergence conditions are reached; The training process uses the validation set to evaluate the model and adjust hyperparameters to optimize model performance.
[0009] The present invention has the following advantages: By improving the compression model and dictionary model method for detecting pipeline leakage signals provided herein, the present invention has the following improvements compared with similar devices: The compression model and dictionary model method for detecting pipeline leakage signals described in the present invention, the multi-layer single dictionary model has great advantages in the reconstruction accuracy, generalization and robustness of leakage signals. Compared with other reconstruction models, it is more suitable for the reconstruction of leakage signals, and provides an innovative compression and reconstruction method for actual leakage monitoring. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] Figure 1 is a flow chart of a multi-layer single dictionary learning model of the present invention; Figure 2 is a model training flow chart of the present invention; Figure 3 is a deep dictionary model test graph of the present invention; Figure 4 is a reconstruction result diagram of the present invention; Figure 5 It is a single dictionary learning schematic diagram of the present invention; Figure 6 It is a schematic diagram of two-layer dictionary learning of the present invention; Figure 7 It is a schematic diagram of the training of the three-layer single dictionary reconstruction model of the present invention; Figure 8 It is the pseudo code for training the multi-layer single dictionary model of the present invention. DETAILED DESCRIPTION
[0011] The following will be combined with the attached Figure 1-Figure 8 The present invention is described in detail, and the technical solutions in the embodiments of the present invention are made clear. The complete description is obvious. The described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0012] It should be noted that when a component is referred to as being "fixed to" another component, it may be directly on the other component or there may also be a component centered. When a component is considered to be "connected to" another component, it may be directly connected to the other component or there may also be a component centered. When a component is considered to be "set on" another component, it may be directly set on the other component or there may also be a component centered. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.
[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.
[0014] See also Figures 1 to 8 , the compression model and dictionary model method for detecting pipeline water leakage signals of the present invention; First, collect diverse data and conduct multi-layer single dictionary model learning on the computer to obtain the compression model and dictionary model; Transplant the compression model at the terminal node to compress and sample the collected signal; The dictionary model is transplanted at the sink node to reconstruct the received compressed signal.
[0015] The architecture of the deep dictionary model includes multiple dictionary layers, each of which consists of a dictionary and a nonlinear activation function; The input data is first encoded by the first dictionary learning layer and then passed to the next dictionary learning layer, and so on until it reaches the last dictionary learning layer.
[0016] The output of the last dictionary learning layer is a sparse representation of the data; The layer-by-layer training process of deep dictionary learning is as follows: First, train and learn the feature representation Z1 and dictionary D1 of the first layer: X=D1×Z1 Then, the feature Z1 of the first layer is decomposed to solve the dictionary D2 and feature Z2 of the second layer: Z1=D2×Z2 Multiple single dictionaries are connected together through a deep learning framework to form a multi-layer single dictionary model; The synthesis framework of single dictionary learning enables signal features to synthesize data along the direction of the dictionary. The principle is shown in the formula; X=D1Z in X Represents input data, D1 represents the dictionary to be learned, Z yes X In the dictionary D1 represents sparse coefficients; The single dictionary learning model is extended to a two-layer dictionary learning model, as shown in the formula; X=D1D2Z2 Where X represents the input data, D1 represents the first-level dictionary to be learned,D2 Represents the second-layer dictionary to be learned. Lines of the same color represent atoms in the same column of the current-layer dictionary. Z2 It is the sparse coefficient of X in the two-layer dictionary; The dictionary is obtained by adopting a greedy iterative learning method, and each layer is learned separately to ensure convergence until the third layer dictionary training is completed; The training process of the three-layer dictionary model is as follows Figure 7 Shown Figure 4 The W1, W2, ..., Wn in the model represent the weight parameters used to sparse the data, that is, the compression model in the sampling node. D1, D2, ..., Dn represent the weight parameters used for reconstruction in the model, that is, the dictionary model in the aggregation node. Z1, Z2, ..., Zn represent the hidden layers of each layer in the training process, that is, the sparse coefficients, and the sparse coefficients of the previous layer serve as the input of the sparse coefficients of the next layer.
[0017] The training process of deep dictionary models usually adopts layer-by-layer training; First, initialize the parameters of the first dictionary layer, including the dictionary and nonlinear activation function; Then, training is performed layer by layer, with each layer encoding the input data into a sparse representation through forward propagation and updating the parameters through backpropagation; After each layer is trained, the learned features are used as the input of the next layer, and the above steps are repeated until the required number of layers or convergence conditions are reached; The training process uses the validation set to evaluate the model and adjust hyperparameters to optimize model performance.
[0018] Finally, the trained dictionary model can be used in various practical applications, such as image processing, speech processing, and natural language processing, to extract important features and structural information of the data and solve practical problems. Figure 2 shown.
[0019] The testing process of the deep dictionary model usually includes the following steps: First, the test data is input into the trained deep dictionary model for forward propagation to obtain a sparse coding representation of the data; These sparse coded representations are then used to perform corresponding tasks such as classification, regression, or generation.
[0020] After the task is completed, the model performance is evaluated according to the task evaluation indicators, including accuracy, mean square error, etc. Based on the evaluation results, the model parameters can be adjusted or other strategies can be adopted to improve the model performance. Finally, the obtained model can be used in practical applications to solve related problems after verification by the test process. Model testing process Figure 3 Shown: (1) Test data preparation; The user prepares the data set for testing, including input data and label data, and reads the original test data from the CSV file; Model loading; Loading the trained model parameters. During the model testing process, loading the trained model is crucial for water leakage signal processing. It allows us to use the features and rules that the model has previously learned from a large amount of water leakage signal data to accurately predict and classify new and unknown water leakage signals, thereby verifying the model's generalization ability and actual application effect.
[0021] (3) Create a data loader The data loader can efficiently prepare and load test data, ensuring that the data is input into the model in a preset manner (such as batches, shuffled, etc.) to accurately evaluate the performance and generalization ability of the model.
[0022] (4) Model inference The role of model inference for water leakage signals is to use the trained model to predict and analyze new, unknown water leakage signals, so as to evaluate the performance of the model in actual application scenarios, determine whether there is a water leakage event, and may further provide a prediction of the location or severity of the leakage.
[0023] (5) Evaluate model performance When evaluating the performance of the model for water leakage signals, MSE (mean square error) and cosine similarity each play a key role. MSE measures the error between the model prediction value and the actual water leakage signal value. The smaller the value, the more accurate the prediction, which helps to ensure the accuracy of the model in quantitative prediction. Cosine similarity is used to measure the similarity between two vectors (such as water leakage signal feature vectors). It is suitable for identifying water leakage signal patterns or performing similarity comparisons, and provides strong support for the classification and recognition capabilities of the model. The combination of the two can comprehensively evaluate the performance of the model in water leakage signal processing.
[0024] (6) Visualizing the reconstruction results Visual reconstruction results play a key role in water leakage signal processing. They can intuitively demonstrate the model's ability to extract and reconstruct water leakage signal features. By comparing the original signal and the reconstructed signal, researchers can evaluate the accuracy of the model in capturing and expressing the key features of water leakage signals, and then optimize the model design or improve the data processing method. This process not only enhances the interpretability of the research results, but also helps to improve the efficiency and accuracy of water leakage signal detection.
[0025] The source of the experimental data set: In real life, a piezoelectric sensor is used to connect a Raspberry Pi 4b to collect leakage data on PVC pipes, including 60,000 pieces of non-leakage data and 220,000 pieces of leakage data. Considering the actual application of the model, the leakage data are divided into 120,000 pieces of small leakage data, 60,000 pieces of medium leakage data, and 40,000 pieces of large leakage data according to the leakage situation. The data set is divided into training samples and test samples with a ratio of 8:2.
[0026] The above shows and describes the basic principle, main features and advantages of the present invention, and the standard parts used in the present invention can be purchased from the market, and special-shaped parts can be customized according to the description in the specification and the drawings. The specific connection methods of each part adopt the conventional means such as mature bolts, rivets, welding, etc. in the prior art. The machinery, parts and equipment all adopt the conventional models in the prior art, and the circuit connection adopts the conventional connection method in the prior art, which will not be described in detail here.
[0027] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A compression model and dictionary model method for detecting pipeline leakage signals, characterized in that: The steps include: Collect diverse data and conduct multi-layer single dictionary model learning on the computer to obtain compression model and dictionary model; Transplant the compression model at the terminal node to compress and sample the collected signal; The dictionary model is transplanted at the sink node to reconstruct the received compressed signal.
2. The compression model and dictionary model method for detecting pipeline leakage signals according to claim 1, characterized in that: Multiple dictionary layers form the architecture of the deep dictionary model, and each dictionary learning layer consists of a dictionary and a nonlinear activation function; The input data is first encoded by the first dictionary learning layer, then passed to the next dictionary learning layer, and so on until it reaches the last dictionary learning layer; The output of the last dictionary learning layer is a sparse representation of the data; The layer-by-layer training process of deep dictionary learning is as follows: First, train and learn the feature representation Z1 and dictionary D1 of the first layer: X=D1×Z1; Then, the feature Z1 of the first layer is decomposed to obtain the dictionary D2 and feature Z2 of the second layer: Z1=D2×Z2.
3. The compression model and dictionary model method for detecting pipeline leakage signals according to claim 2, characterized in that: Multiple single dictionaries are connected together through a deep learning framework to form a multi-layer single dictionary model; The synthesis framework of single dictionary learning enables signal features to synthesize data along the direction of the dictionary. The principle is shown in the formula; X=D1Z in X Represents input data, D1 represents the dictionary to be learned, Z yes X In the dictionary D1 represents sparse coefficients; The single dictionary learning model is extended to a two-layer dictionary learning model, as shown in the formula; X=D1D2Z2 Where X represents the input data, D1 represents the first-level dictionary to be learned, D2 Represents the second-layer dictionary to be learned. Lines of the same color represent atoms in the same column of the current-layer dictionary. Z2 It is the sparse coefficient of X in the two-layer dictionary; The dictionary is obtained by adopting a greedy iterative learning method, and each layer is learned separately to ensure convergence until the third layer dictionary training is completed.
4. The compression model and dictionary model method for detecting pipeline water leakage signals according to claim 3, characterized in that: The layer-by-layer training of the deep dictionary model includes the following steps: First, initialize the parameters of the first dictionary layer, including the dictionary and nonlinear activation function; Then, training is performed layer by layer, with each layer encoding the input data into a sparse representation through forward propagation and updating the parameters through backpropagation; After each layer is trained, the learned features are used as the input of the next layer, and the above steps are repeated until the required number of layers or convergence conditions are reached; The training process uses the validation set to evaluate the model and adjust hyperparameters to optimize model performance.