Automatic analysis system and method for abnormal state of pressure gauge

Abstract

The invention discloses a system and a method for automatically analyzing the abnormal state of a pressure gauge, and the method comprises the steps: learning a cooperation mode between a real-time flow velocity value sequence and a real-time pressure value sequence under a normal working condition through a cooperation model based on an auto-encoder, and reasoning an expected real-time pressure value sequence according to the real-time flow velocity value sequence. By calculating the residual error between the actual real-time pressure value sequence and the reasoning real-time pressure value sequence, the deviation degree of the cooperative relationship between the pressure and the flow velocity can be effectively quantified. And the residual error sequence and the time sequence characteristics of the residual error sequence are combined with the pressure time sequence characteristics of the target pressure gauge and are jointly input into the trained abnormal state diagnosis multi-classifier model, so that the accurate and robust diagnosis of the abnormal state of the pressure gauge is realized, and the pressure change caused by the fault of the pressure gauge or the abnormal system process is effectively distinguished.

Description

Technical Field

[0001] This application relates to the field of intelligent analysis, and more specifically, to an automatic analysis system and method for abnormal states of pressure gauges. Background Art

[0002] In the process of modern industrial production and pipeline network operation, pressure gauges, as key process monitoring instruments, play an important role in ensuring the safe and stable operation of the system. Since abnormal pressure often indicates equipment failures, pipeline leaks, or other potential risks, timely detection and accurate diagnosis of the abnormal state of pressure gauges are of great significance for preventing accidents, reducing losses, and improving operation and maintenance efficiency. With the continuous improvement of automation and intelligence levels, it has become difficult to meet the actual needs of efficient and accurate abnormal detection by relying on manual inspections or single-threshold alarms. Therefore, there is an urgent need to construct a new technical solution that can automatically analyze the abnormal state of pressure gauges.

[0003] Currently, most existing automatic analysis solutions for abnormal states of pressure gauges mainly rely on the trend judgment of single parameters. For example, simple alarms are achieved by setting upper and lower pressure thresholds. However, these existing methods generally have limitations. They often only rely on single pressure data for analysis and do not fully consider other relevant parameters in the system environment where the pressure gauge is located, such as flow rate. This single-dimensional analysis method makes it difficult to effectively distinguish whether the pressure change is caused by the failure of the pressure gauge itself (such as sensor drift, blockage) or system process abnormalities (such as pipeline leaks, pump failures) in the face of complex and changing industrial scenarios. As a result, the accuracy, robustness, and specificity of abnormal diagnosis are insufficient, especially in distinguishing sensor failures from process failures, which is prone to false alarms or missed alarms and cannot meet the industrial diagnosis requirements of high precision and high reliability.

[0004] Therefore, an optimized automatic analysis solution for abnormal states of pressure gauges is expected. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. The embodiments of this application provide an automatic analysis system and method for abnormal states of pressure gauges. By using a collaborative model based on an autoencoder, it learns the collaborative pattern between the real-time flow rate value sequence and the real-time pressure value sequence under normal working conditions, and accordingly infers the expected real-time pressure value sequence from the real-time flow rate value sequence. By calculating the residual between the actual real-time pressure value sequence and the inferred real-time pressure value sequence, the deviation degree of the collaborative relationship between pressure and flow rate can be effectively quantified. This residual sequence and its temporal characteristics, combined with the pressure temporal characteristics of the target pressure gauge itself, are jointly input into the trained multi-classifier model for abnormal state diagnosis, so as to achieve accurate and robust diagnosis of the abnormal state of the pressure gauge and effectively distinguish whether the pressure change is caused by the failure of the pressure gauge itself or system process abnormalities.

[0006] According to one aspect of the present application, an automatic analysis method for abnormal states of a pressure gauge is provided, which includes: Obtaining a real-time pressure value sequence of a target pressure gauge; Obtaining a real-time flow rate value sequence collected by a flowmeter associated with the target pressure gauge; Performing timestamp alignment on the real-time pressure value sequence and the real-time flow rate value sequence to obtain an aligned real-time pressure value sequence and an aligned real-time flow rate value sequence; Performing individual signal feature extraction on the aligned real-time pressure value sequence to obtain a pressure time series feature vector of the target pressure gauge; Inputting the aligned real-time flow rate value sequence into a collaborative model based on an autoencoder to obtain an inferred real-time pressure value sequence; Calculating a pressure-flow rate collaborative deviation feature between the inferred real-time pressure value sequence and the real-time pressure value sequence, and inputting the pressure-flow rate collaborative deviation feature and the pressure time series feature of the target pressure gauge into a trained abnormal state diagnosis multi-classifier model to obtain an abnormal diagnosis result.

[0007] According to another aspect of the present application, an automatic analysis system for abnormal states of a pressure gauge is provided, which includes: A pressure value acquisition module for obtaining a real-time pressure value sequence of a target pressure gauge; A flow rate value acquisition module for obtaining a real-time flow rate value sequence collected by a flowmeter associated with the target pressure gauge; A timestamp alignment module for performing timestamp alignment on the real-time pressure value sequence and the real-time flow rate value sequence to obtain an aligned real-time pressure value sequence and an aligned real-time flow rate value sequence; An individual signal feature extraction module for performing individual signal feature extraction on the aligned real-time pressure value sequence to obtain a pressure time series feature vector of the target pressure gauge; A pressure value inference module for inputting the aligned real-time flow rate value sequence into a collaborative model based on an autoencoder to obtain an inferred real-time pressure value sequence; An abnormal state diagnosis module for calculating a pressure-flow rate collaborative deviation feature between the inferred real-time pressure value sequence and the real-time pressure value sequence, and inputting the pressure-flow rate collaborative deviation feature and the pressure time series feature of the target pressure gauge into a trained abnormal state diagnosis multi-classifier model to obtain an abnormal diagnosis result.

[0008] Compared with the prior art, an automatic analysis system and method for abnormal states of a pressure gauge provided by this application utilize a collaborative model based on an autoencoder to learn the collaborative pattern between the real-time flow rate value sequence and the real-time pressure value sequence under normal operating conditions, and accordingly infer the expected real-time pressure value sequence from the real-time flow rate value sequence. By calculating the residual between the actual real-time pressure value sequence and the inferred real-time pressure value sequence, the deviation degree of the collaborative relationship between pressure and flow rate can be effectively quantified. This residual sequence and its temporal characteristics, combined with the pressure temporal characteristics of the target pressure gauge itself, are jointly input into the trained multi-classifier model for abnormal state diagnosis, thereby achieving accurate and robust diagnosis of the abnormal state of the pressure gauge and effectively distinguishing whether the pressure change is caused by a fault of the pressure gauge itself or an abnormality in the system process. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of this application in more detail with reference to the accompanying drawings, the above and other objects, features, and advantages of this application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of this application, and constitute a part of the specification. Together with the embodiments of this application, they are used to explain this application, and do not constitute a limitation to this application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 It is a flowchart of the automatic analysis method for abnormal states of a pressure gauge according to an embodiment of this application; Figure 2 It is a schematic diagram of data flow of the automatic analysis method for abnormal states of a pressure gauge according to an embodiment of this application; Figure 3 It is a flowchart of sub-step S6 of the automatic analysis method for abnormal states of a pressure gauge according to an embodiment of this application; Figure 4 It is a block diagram of the automatic analysis system for abnormal states of a pressure gauge according to an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0011] Next, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments of this application. It should be understood that this application is not limited by the exemplary embodiments described herein.

[0012] As shown in this application and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one", and / or "the" do not specifically refer to the singular, but may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0013] Although this application makes various references to certain modules in the system according to embodiments of this application, any number of different modules may be used and run on a user terminal and / or a server. The modules are merely illustrative, and different aspects of the system and method may use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the operations above or below do not necessarily have to be performed precisely in order. Instead, various steps may be processed in reverse order or simultaneously as needed. Also, other operations may be added to these processes, or one or more steps may be removed from these processes.

[0015] Next, example embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all embodiments of this application. It should be understood that this application is not limited by the example embodiments described herein.

[0016] In the technical solution of this application, a method for automatically analyzing the abnormal state of a pressure gauge is proposed. Figure 1 It is a flowchart of the method for automatically analyzing the abnormal state of a pressure gauge according to an embodiment of this application. Figure 2 It is a schematic diagram of data flow of the method for automatically analyzing the abnormal state of a pressure gauge according to an embodiment of this application. As Figure 1 and Figure 2 shown, the method for automatically analyzing the abnormal state of a pressure gauge according to an embodiment of this application includes the steps of: S1, obtaining a real-time pressure value sequence of a target pressure gauge; S2, obtaining a real-time flow velocity value sequence collected by a flowmeter associated with the target pressure gauge; S3, performing timestamp alignment on the real-time pressure value sequence and the real-time flow velocity value sequence to obtain an aligned real-time pressure value sequence and an aligned real-time flow velocity value sequence; S4, extracting individual signal features from the aligned real-time pressure value sequence to obtain a pressure time series feature vector of the target pressure gauge; S5, inputting the aligned real-time flow velocity value sequence into a collaborative model based on an autoencoder to obtain an inferred real-time pressure value sequence; S6, calculating a pressure-flow velocity collaborative deviation feature between the inferred real-time pressure value sequence and the real-time pressure value sequence, and inputting the pressure-flow velocity collaborative deviation feature and the pressure time series feature of the target pressure gauge into a trained abnormal state diagnosis multi-classifier model to obtain an abnormal diagnosis result.

[0017] Specifically, in step S1, obtain the real-time pressure value sequence of the target pressure gauge; and obtain the real-time flow rate value sequence collected by the flow meter associated with the target pressure gauge. It should be understood that it is often difficult for the data of a single parameter to accurately reflect the true state of equipment operation under complex working conditions, while multi-source data fusion can reveal deeper collaborative relationships between parameters. For example, under normal circumstances, there should be a certain physical correlation between the flow rate and the pressure. When a pipeline leaks, becomes blocked, or the instrument itself malfunctions, this correlation will be broken, resulting in abnormal characteristics in the data between the two. Therefore, by simultaneously obtaining and aligning these two types of time series data, a complete and accurate information basis can be provided for subsequent model analysis.

[0018] Specifically, in step S3, perform timestamp alignment on the real-time pressure value sequence and the real-time flow rate value sequence to obtain the aligned real-time pressure value sequence and the aligned real-time flow rate value sequence. Due to the influence of factors such as sampling frequency, communication delay, and device clock error in the actual operating environment, the pressure value sequence and the flow rate value sequence often have time asynchrony. If the original data is directly input into the subsequent analysis model, it is easy to weaken or even lose the correlation between parameters, thereby affecting the accuracy of anomaly detection. Therefore, in the technical solution of this application, timestamp alignment is performed on the real-time pressure value sequence and the real-time flow rate value sequence to obtain the aligned real-time pressure value sequence and the aligned real-time flow rate value sequence. This not only helps the model accurately mine the dynamic correlation between parameters but also provides a prerequisite for intelligent algorithms such as residual analysis and autoencoder inference. Through the aligned data, complex abnormal situations that are difficult to detect by single variables alone can be more effectively identified, realizing efficient fault diagnosis driven by multi-parameter collaboration. In a specific example of this application, a target time axis is constructed with a step size of seconds. For each target moment, the data point closest to this moment is found from the original sequence and numerical completion is performed. For data with a large number of missing values or extremely uneven sampling, smoothing processing can also be performed by means of sliding window weighted averaging, etc. This step significantly improves the sensitivity and robustness of the anomaly detection system to the dynamic changes of the process, laying a solid foundation for realizing intelligent and automated operation and maintenance management.

[0019] Specifically, in step S4, individual signal feature extraction is performed on the aligned real-time pressure value sequence to obtain the pressure time-series feature vector of the target pressure gauge. It should be understood that the pressure data collected by the pressure gauge is often affected by various working conditions, presenting complex time-series dynamic characteristics. Therefore, in the technical solution of this application, in order to achieve efficient and accurate identification of the abnormal state of the pressure gauge, after the time stamps of the real-time pressure value sequence and the flow velocity value sequence are aligned, individual signal feature extraction is further performed on the aligned real-time pressure value sequence to fully explore and express the change law, fluctuation pattern and potential abnormal trend of the target pressure gauge itself in the historical period, so as to obtain the pressure time-series feature vector of the target pressure gauge. Among them, the pressure time-series feature vector of the target pressure gauge not only contains the trend information of the pressure signal changing with time, but also can capture key dynamic attributes such as periodic fluctuations, mutation points and abnormal persistence, thus effectively improving the discrimination ability of the subsequent multi-classifier model for different types of abnormal states (such as leakage, blockage or instrument failure, etc.). In a specific example of this application, individual signal feature extraction based on LSTM is performed on the aligned real-time pressure value sequence to obtain the pressure time-series feature vector of the target pressure gauge. Here, as a recurrent neural network specialized in processing time-series data dependencies, LSTM can effectively model the complex correlations between the current pressure value and historical multi-step data. By performing individual signal feature extraction on the aligned real-time pressure value sequence based on the LSTM model, not only the information expression ability of the original data is greatly enriched, but also solid data support is provided for the entire intelligent diagnosis system.

[0020] Specifically, in step S5, the aligned real-time flow velocity value sequence is input into a collaborative model based on an autoencoder to obtain an inferred real-time pressure value sequence. It should be understood that there is a close physical coupling relationship between pressure and flow velocity, and traditional methods often ignore the complex and implicit coupling mechanism between different parameters. When the system is in a normal state, the change in flow velocity usually affects the change in pressure in a certain pattern, while when abnormalities such as leakage, blockage or instrument failure occur, this pattern will be broken. Therefore, in the technical solution of this application, the aligned real-time flow velocity value sequence is input into a collaborative model based on an autoencoder to obtain an inferred real-time pressure value sequence. Here, the autoencoder is an unsupervised learning neural network structure mainly used to automatically learn the effective features in the input data, and it consists of two parts: an encoder that compresses the input signal into a low-dimensional latent space representation, and a decoder that attempts to reconstruct the original input from this latent representation. In the specific implementation process, the complex mapping collaborative relationship between flow velocity and pressure is captured by the autoencoder, and then the theoretical pressure value sequence inferred at the corresponding moment is output by the decoding end.

[0021] Specifically, in step S6, the pressure-flow rate collaborative deviation feature between the inferred real-time pressure value sequence and the real-time pressure value sequence is calculated, and the pressure-flow rate collaborative deviation feature and the target pressure gauge pressure time series feature are input into the trained multi-classifier model for abnormal state diagnosis to obtain an abnormal diagnosis result. In a specific example of the present application, as Figure 3 shown, step S6 includes: S61, calculating the residual between the inferred real-time pressure value sequence and the real-time pressure value sequence to obtain a real-time pressure value residual sequence; S62, extracting the residual time series feature from the real-time pressure value residual sequence to obtain a pressure-flow rate collaborative deviation feature vector as the pressure-flow rate collaborative deviation feature; S63, performing multi-scale joint progressive perception on the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure time series feature vector to obtain a target pressure gauge main body behavior - collaborative behavior time series joint coding vector; S64, obtaining an abnormal diagnosis result based on the target pressure gauge main body behavior - collaborative behavior time series joint coding vector.

[0022] Specifically, in step S61, the residual between the inferred real-time pressure value sequence and the real-time pressure value sequence is calculated to obtain a real-time pressure value residual sequence. It should be understood that through residual analysis, complex anomalies that are difficult to identify or easily overlooked based on the trend of a single parameter can be effectively captured. Even if some anomalies do not cause the absolute value to exceed the limit, as long as they disrupt the normal collaborative relationship between the flow rate and pressure, they will be manifested as significant fluctuations or continuous offsets in the residual. For example, when there is a small leak in the pipeline, the actual pressure will be lower than the model inference value, and instrument failure may cause a persistent deviation between the two. Therefore, in the technical solution of the present application, the inferred real-time pressure value sequence is compared with the real-time pressure value sequence actually observed, and the residual between the two is calculated to effectively capture complex anomalies that are difficult to identify based on the trend of a single parameter.

[0023] In specific implementation, for each moment, the actually collected pressure value is subtracted from the theoretical pressure value predicted by the autoencoder model based on the flow rate to obtain the residual at that moment; similarly, after traversing each time window, a real-time pressure value residual sequence is obtained. The real-time pressure value residual sequence not only intuitively reflects the system operation state and health level at each stage of the process, but also provides a rich information basis for subsequent feature extraction and multi-class diagnosis. By analyzing this residual sequence, various small but critical data deviations can be keenly captured, achieving a more accurate and robust automated anomaly detection effect, and greatly improving the fine-grained anomaly recognition ability of traditional monitoring means for complex working conditions, providing solid data support and technical guarantee for intelligent operation and maintenance and risk prevention and control in industrial sites.

[0024] Specifically, in S62, residual time-series feature extraction is performed on the real-time pressure value residual sequence to obtain a pressure-flow rate collaborative deviation feature vector as the pressure-flow rate collaborative deviation feature. It should be understood that the simple residual value can only reveal the deviation at a certain moment, and cannot comprehensively reflect the evolution law and persistence of the abnormal state in the time dimension. Industrial working conditions are often complex and changeable, and some abnormalities (such as leaks, blockages, or sensor drifts) have obvious time-series characteristics and dynamic evolution processes. Therefore, in the technical solution of this application, residual time-series feature extraction is performed on the real-time pressure value residual sequence to convert the original high-dimensional, redundant, and noisy residual signal into a highly generalized low-dimensional vector, enabling it to fully express key information such as the dynamic change trend, periodic fluctuation, mutation point, and persistent offset shown by the flow rate-pressure collaborative disorder during the system operation process, and obtaining a pressure-flow rate collaborative deviation feature vector.

[0025] In a specific example of this application, a gated recurrent unit (GRU) network can be used as the main residual time-series feature extraction model. The GRU is a recurrent neural network structure that can efficiently process long-term dependence relationships and effectively suppress the problem of gradient disappearance, and is very suitable for modeling non-linear dynamic processes in industrial data. During this process, the fully aligned real-time pressure value residual sequence is input into the GRU network. The network automatically learns and filters out important historical information related to abnormalities through the gating mechanism, gradually encodes the input signal, and generates a fixed-length, highly expressive, and discriminative pressure-flow rate collaborative deviation feature vector at the output end.

[0026] Specifically, the S63 performs multi-scale joint progressive perception on the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure time series feature vector to obtain the target pressure gauge main behavior-collaborative behavior time series joint coding vector. It should be understood that the dynamic collaborative relationship between pressure and flow rate has a multi-level and multi-granular complex coupling characteristic. For example, when a leak occurs in the pipeline network, the linkage relationship between pressure and flow rate may appear as a superposition of short-term sudden local anomalies (such as a sudden drop in pressure accompanied by a sharp increase in flow rate) and long-term trend deviations (such as pressure continuously below the normal range); and when the pressure gauge itself has a drift failure, its pressure time series may appear as a steady-state offset, but the collaborative deviation with the flow rate remains normal. The traditional single-layer fusion method cannot distinguish these abnormal modes at different abstract levels. Therefore, in the technical solution of the present application, the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure time series feature vector are multi-scale joint progressively perceived to obtain the target pressure gauge main behavior-collaborative behavior time series joint coding vector. That is, by decoupling the fine-grained differences between the main behavior (its own timing characteristics) and the coordinated behavior (dynamic relationship between pressure and flow rate) of the pressure gauge in a hierarchical manner, the essential characteristics of the fault can be captured more accurately. Specifically, through deep nonlinear transformation and cross-level feature interaction, a joint semantic space is constructed that can not only reflect the operating status of the pressure gauge itself (such as individual behaviors such as periodic fluctuations and trend drift), but also model its dynamic coordinated relationship with flow rate parameters (such as the break in pressure-flow rate correlation caused by leakage, and the coordinated response delay caused by blockage, etc.). For example, in low-level feature fusion, the system may capture the correspondence between instantaneous pressure jitter and high-frequency flow rate noise to eliminate environmental interference; in mid-level fusion, the coordinated response mode of pressure and flow rate in specific process stages (such as pump start and stop, valve adjustment) can be identified to determine whether there is a linkage anomaly; and in deep-level fusion, instrument failure or systemic failure is identified through global reasoning at the semantic level (such as the contradiction between the long-term trend of pressure and the accumulated flow rate). This hierarchical and progressive perception mechanism enables the system to cover all aspects from local details of the time series to global semantics, avoiding pattern omissions caused by single-scale analysis, thereby achieving high-precision fault type discrimination in the classifier (such as distinguishing leaks, blockages, or instrument failures), and reducing the risk of misjudgment caused by single-dimensional analysis or simple feature splicing in traditional methods.

[0027] Specifically, first, multi-level implicit feature extraction is performed on the pressure-flow rate co-deviation feature vector and the target pressure gauge pressure time series feature vector to obtain the middle-level implicit feature encoding vector of the pressure-flow rate co-deviation, the middle-level time series implicit feature encoding vector of the target pressure gauge pressure, the deep-level implicit feature encoding vector of the pressure-flow rate co-deviation, and the deep-level time series implicit feature encoding vector of the target pressure gauge pressure. It should be understood that there are multi-dimensional abnormal representations in the dynamic relationship between pressure and flow rate, and traditional single-layer feature extraction cannot capture such cross-level contradictory signals. Therefore, in the technical solution of this application, multi-level implicit feature extraction is performed on the pressure-flow rate co-deviation feature vector and the target pressure gauge pressure time series feature vector to obtain the middle-level implicit feature encoding vector of the pressure-flow rate co-deviation, the middle-level time series implicit feature encoding vector of the target pressure gauge pressure, the deep-level implicit feature encoding vector of the pressure-flow rate co-deviation, and the deep-level time series implicit feature encoding vector of the target pressure gauge pressure.

[0028] Specifically, the middle-level implicit feature encoding vector focuses on the structural features of the pressure-flow rate co-deviation (such as local waveform distortion, phase lag within a specific time window, etc.), while the deep-level implicit feature encoding vector is dedicated to capturing the global semantics of the individual behavior and collaborative behavior of the pressure gauge (such as the contradiction between the long-term deviation of pressure from the normal range and the cumulative value of flow rate). This hierarchical abstraction mechanism enables the system to not only focus on local details (such as the matching degree between a certain pressure pulse and flow rate response), but also take into account the overall trend (such as the relationship between the mean pressure shift and flow rate distribution), ultimately forming a multi-scale fault recognition framework and significantly improving the separability of abnormal patterns.

[0029] In a specific example of this application, the following formula is used to perform multi-level implicit feature extraction on the pressure-flow rate co-deviation feature vector and the target pressure gauge pressure time series feature vector to obtain the middle-level implicit feature encoding vector of the pressure-flow rate co-deviation, the middle-level time series implicit feature encoding vector of the target pressure gauge pressure, the deep-level implicit feature encoding vector of the pressure-flow rate co-deviation, and the deep-level time series implicit feature encoding vector of the target pressure gauge pressure; where the formula is:

[0030] Where, is the pressure-flow rate co-deviation feature vector, is the target pressure gauge pressure time series feature vector, represents the weight matrix for middle-level feature extraction, represents the bias term for middle-level feature extraction, represents function, is the middle-level implicit feature encoding vector of the pressure-flow rate co-deviation, is the middle-level time series implicit feature encoding vector of the target pressure gauge pressure, represents the weight matrix for deep feature extraction, represents the bias term for deep feature extraction, is the deep implicit feature encoding vector of the pressure-velocity synergistic deviation, The deep temporal implicit feature encoding vector of the target pressure gauge pressure.

[0031] Next, the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure time series feature vector are fused at a low level to obtain the target pressure gauge main behavior-collaborative behavior low-level time series fusion feature encoding vector. It should be understood that the time series data collected by the pressure gauge and the flow meter have highly dynamic coupling characteristics. For example, when the initial leakage caused by a small crack in the pipeline network occurs, the collaborative deviation of pressure and flow rate may be manifested as the precise timestamp misalignment of the pressure drop and the flow rate surge on the millisecond time scale. If the middle and high-level abstract features are directly used for interaction, it is easy to lose the fine time correlation of the original time series signal due to the layer-by-layer abstraction characteristics of the deep neural network. Low-level feature fusion can retain the millisecond-level synchronization characteristics of the pressure pulse waveform and the flow rate response by directly operating the fine structure of the original time series data, providing a data basis with clear physical meaning for identifying such transient anomalies. By constructing this fusion base at the original signal level, the system not only enhances the early recognition capability of complex anomalies such as micro-leakage and intermittent blockage, but also ensures the interpretability of the physical mechanism in the multi-scale feature evolution process.

[0032] In a specific example of the present application, the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure time series feature vector are subjected to low-level feature fusion using the following formula to obtain the target pressure gauge main behavior-collaborative behavior low-level time series fusion feature encoding vector; wherein the formula is:

[0033] in, It means adding by position. represents a multi-layer perceptron, A low-level temporal fusion feature encoding vector of the main behavior-cooperative behavior of the target pressure gauge is provided.

[0034] Subsequently, perform mid-level feature fusion on the mid-level implicit feature encoding vector in the pressure-flow rate coordination deviation and the mid-level temporal implicit feature encoding vector of the target pressure gauge pressure to obtain the mid-level temporal fusion feature encoding vector of the target pressure gauge main body behavior - coordination behavior. It should be understood that when a certain regulating valve in the pipe network system has a jamming fault, the mid-level temporal characteristics of the pressure gauge may be manifested as a phase shift of the periodic pressure fluctuation, while the corresponding mid-level characteristics of the flow rate coordination deviation may capture the step-like mutation of the flow rate adjustment curve. Such cross-modal structural anomalies may be masked by noise at the original signal level, but after mid-level feature extraction, the component-level behavior patterns of both (such as the correlation between valve opening and flow resistance) become explicit.

[0035] Compared with the instantaneous signal correspondence focused on by low-level fusion, mid-level fusion focuses on the structured pattern matching within the equipment operation cycle. For example, in the scenario of early wear of the centrifugal pump bearing, the mid-level characteristics of the pressure gauge may extract the harmonic distortion of the impeller passing frequency, while the mid-level characteristics of the flow rate coordination deviation may identify the abnormal modulation of the flow rate pulsation amplitude. Through mid-level feature fusion, a causal relationship model between the two on the rotating machinery vibration transmission path can be established. This fusion mechanism enables the system to break through the limitations of a single sensor perspective, starting from the essence of physical interaction between equipment components, and analyzing the abnormal measurement value of the pressure gauge and the response mode of the flow rate parameter in a unified mechanical system model, so as to distinguish transient anomalies caused by external disturbances from structural faults caused by actual equipment deterioration. This structured feature fusion based on the equipment operation mechanism not only enhances the system's ability to analyze multiple fault concurrent scenarios, but more importantly, by capturing the stable correlation pattern between pressure and flow rate at the component-level behavior level, it significantly reduces the misjudgment probability caused by random noise or instantaneous disturbances, providing a more interpretable diagnostic basis for operation and maintenance decisions.

[0036] In a specific example of the present application, the following formula is used to perform mid-level feature fusion on the mid-level implicit feature encoding vector of the pressure-flow rate coordination deviation and the mid-level temporal implicit feature encoding vector of the target pressure gauge pressure to obtain the mid-level temporal fusion feature encoding vector of the target pressure gauge main body behavior - coordination behavior; where the formula is:

[0037] Where, represents element-wise multiplication by position, represents the attention mechanism, represents function, represents the feature dimension of the vector, is the mid-level temporal fusion feature encoding vector of the target pressure gauge main body behavior - coordination behavior.

[0038] Then, deep feature fusion is performed on the deep implicit feature encoding vector of the pressure-flow rate collaborative deviation and the deep temporal implicit feature encoding vector of the target pressure gauge pressure to obtain the deep-level temporal fusion feature encoding vector of the target pressure gauge main body behavior - collaborative behavior. It should be understood that when the pipe network system undergoes structural aging or regional corrosion, the deep temporal features of the pressure gauge may implicitly encode the entropy increase characteristics of long-term pressure decay, while the deep collaborative deviation features may map the systematic deviation of the pressure-flow rate energy conservation relationship. Such global anomalies are difficult to detect in local temporal segments, but through deep feature extraction, the semantic-level representations of both (such as the degradation of system energy transfer efficiency and the qualitative change of hydrodynamic characteristics) are made explicit. Here, deep fusion places the long-term evolution law of the individual behavior of the pressure gauge (such as the cumulative effect of zero-point drift) and the essential changes in the pressure-flow rate collaborative relationship (such as the anomaly of the friction coefficient in the Darcy formula) within a unified system-level health assessment framework, thereby constructing a causal association model of equipment failure and fluid state change at a higher abstract level.

[0039] Compared with the component-level interaction patterns focused on by mid-level fusion, deep feature fusion focuses on the global operation essence of industrial systems and the collaborative mechanism under the constraints of physical laws. For example, in the scenario of overall elastic degradation of the pipe network, the deep pressure features may capture the change in pressure wave propagation characteristics caused by the attenuation of Young's modulus, while the deep collaborative deviation features may reflect the anomaly of the dependence of the flow rate-pressure drop relationship on pipe stiffness. Through deep fusion, a cross-scale correlation between material property degradation and hydrodynamic response can be established. This fusion mechanism enables the system to break through the dependence on explicit parameters of traditional monitoring methods, associate the measurement anomalies of the pressure gauge with the deep semantic changes of the flow rate parameters as system-level state migrations, and thus identify progressive failure modes that cannot be detected by traditional methods. This fusion method based on the essential laws of the system not only enhances the interpretability of the diagnostic results, but more importantly, by capturing the deep coupling relationship between pressure and flow rate at the physical constraint level, constructs an intelligent judgment benchmark beyond empirical thresholds, providing a new technical path for the resilience assessment and life prediction of industrial systems.

[0040] In a specific example of this application, the following formula is used to perform deep feature fusion on the deep implicit feature encoding vector of the pressure-flow rate collaborative deviation and the deep temporal implicit feature encoding vector of the target pressure gauge pressure to obtain the deep-level temporal fusion feature encoding vector of the target pressure gauge main body behavior - collaborative behavior; where the formula is;

[0041] where, represents the weight matrix of deep fusion, represents layer normalization, represents function, and is a low-rank projection matrix, is the target pressure gauge body behavior - collaborative behavior deep-level time series fusion feature coding vector.

[0042] Furthermore, perform target pressure gauge body behavior - collaborative behavior multi-scale progressive complementary perception fusion on the target pressure gauge body behavior - collaborative behavior low-level time series fusion feature coding vector, the target pressure gauge body behavior - collaborative behavior middle-level time series fusion feature coding vector, and the target pressure gauge body behavior - collaborative behavior deep-level time series fusion feature coding vector to obtain the target pressure gauge body behavior - collaborative behavior time series joint coding vector. It should be understood that the low-level time series fusion feature coding vector carries the millisecond-level correlation details of the original pressure fluctuation and flow velocity collaborative deviation, while the middle-level feature coding vector contains the system behavior patterns that have been preliminarily abstracted. When performing cross-level interaction fusion, direct feature splicing or weighted superposition is likely to cause phase conflicts between hierarchical features. For example, the middle-level abstracted pressure trend feature may obscure the precise time-locking relationship between the sudden pressure pulse and flow velocity oscillation recorded in the low-level features. Therefore, in the preferred example of this application, first, perform cross-level preliminary interaction fusion based on order parameter phase complementarity on the target pressure gauge body behavior - collaborative behavior low-level time series fusion feature coding vector and the target pressure gauge body behavior - collaborative behavior middle-level time series fusion feature coding vector to obtain the target pressure gauge body behavior - collaborative behavior cross-level preliminary fusion semantic feature coding vector. That is, by establishing a dynamic phase compensation channel between the high-frequency details of the low-level features and the macroscopic trends of the middle-level features, key cross-level correlations such as the flow velocity response delay corresponding to a sudden pressure drop are maintained in spatio-temporal consistency. In the oil pipeline leakage detection scenario, the local pressure drop recorded by the low-level features and the abnormal pressure gradient of the entire pipeline network revealed by the middle-level features form a synergistically enhanced signal expression through the constraint of the order parameter field with phase complementarity, which can effectively suppress the interference of transient noise caused by pump valve start-stop operations on feature fusion. This cross-level preliminary semantic fusion not only provides a fusion basis vector with physical field synchronization characteristics for subsequent deep feature interactions, but also maintains the energy conservation characteristics in the multi-scale feature evolution process through the constraint conditions of path integral optimization, ensuring the integrity of the dynamic association of leakage signals during propagation at different abstraction levels, and laying a multi-scale joint perception foundation for accurately distinguishing complex abnormal patterns such as instrument drift and pipe wall microcracks.

[0043] Furthermore, it should be understood that the cross-level preliminary fusion semantic feature encoding vector carries the system behavior patterns formed by the interaction of low and middle-level features, while the deep-level temporal fusion feature encoding vector contains the highly abstracted system dynamic evolution laws. When performing cross-level semantic interaction analysis, direct feature correlation is prone to phase conflicts due to the jump in the abstraction level between levels. For example, there may be a temporal phase misalignment between the overall pressure decay trend of the pipeline network represented in the deep features and the local pressure pulse response recorded in the cross-level preliminary fusion features. Therefore, in the preferred example of this application, a temporal semantic interaction analysis of the target pressure gauge main body behavior - collaborative behavior cross-level preliminary fusion semantic feature encoding vector and the target pressure gauge main body behavior - collaborative behavior deep-level temporal fusion feature encoding vector based on order parameter field constraint and path integral optimization is further performed to obtain the target pressure gauge main body behavior - collaborative behavior temporal joint encoding vector. That is, by constructing a weight matrix field with self-organization characteristics, the energy transfer paths of pressure wave propagation and flow velocity response at different spatio-temporal scales are simulated. The interaction analysis based on order parameter field constraint and path integral optimization significantly improves the system's decoding ability for implicit associated faults. When the dynamic balance of multi-level features under physical constraints is accurately modeled, the system can extract diagnostically explanatory evidence from seemingly discrete abnormal signals. For example, in the compound fault of pressure gauge impulse line blockage accompanied by a change in the pipeline network topology, path integral optimization identifies the coupling effect of sudden increase in local flow resistance and distortion of the global pressure distribution by analyzing the curvature change of the feature manifold, while the order parameter field constraint ensures that this analysis conforms to the flow velocity - pressure drop relationship of Darcy's law, thus effectively distinguishing real blockage from sensor drift.

[0044] In this example, the following formula is used to perform multi-scale progressive complementary perception fusion of the target pressure gauge main body behavior - collaborative behavior low-level temporal fusion feature encoding vector, the target pressure gauge main body behavior - collaborative behavior middle-level temporal fusion feature encoding vector, and the target pressure gauge main body behavior - collaborative behavior deep-level temporal fusion feature encoding vector to obtain the target pressure gauge main body behavior - collaborative behavior temporal joint encoding vector; where the formula is:

[0045] Where, represents the gating weight matrix, represents function, represents the gating order parameter, is the target pressure gauge main body behavior - collaborative behavior cross-level preliminary fusion semantic feature encoding vector, represents the query weight matrix, represents the key weight matrix, represents the value weight matrix, denotes the nuclear norm of the matrix, i.e., the sum of the eigenvalues of the matrix, is the scaling hyperparameter, is the field homogeneous state alignment loss function, denotes the query, denotes the key vector, denotes the value vector, is the target pressure gauge main body behavior - collaborative behavior time - series joint encoding vector.

[0046] Specifically, for S64, an anomaly diagnosis result is obtained based on the target pressure gauge main body behavior - collaborative behavior time - series joint encoding vector. That is, in the technical solution of this application, the target pressure gauge main body behavior - collaborative behavior time - series joint encoding vector is input into the trained anomaly status diagnosis multi - classifier model to obtain the anomaly diagnosis result. It should be understood that the abnormal status of the pressure gauge may be caused by various complex factors, including equipment aging, pipeline leakage, blockage, or sensor failure, etc. It is difficult to accurately identify multiple types of anomalies relying solely on a single parameter or simple rules. Therefore, to achieve efficient discrimination and accurate positioning of different types of pressure gauge abnormal statuses (such as leakage, blockage, sensor drift, etc.), thereby improving the overall intelligent operation and maintenance level and reducing the false alarm rate and missed detection risk, in the technical solution of this application, the deep time - series features reflecting the main body behavior and collaborative behavior during the system operation process are jointly encoded and further input into a multi - classifier model with strong discriminative ability for intelligent diagnosis.

[0047] The multi - classifier model usually adopts classic machine learning or deep - learning structures such as Softmax regression, multi - layer perceptron (MLP), support vector machine (SVM), or ensemble learning methods. Its core working principle is to first model the feature distributions corresponding to various abnormal statuses using the labeled historical sample data, and optimize the parameters through repeated training, so that the model can automatically capture the implicit but key discriminant boundaries between different categories. Then, in the inference stage, taking the target pressure gauge main body behavior - collaborative behavior time - series joint encoding vector obtained by multi - scale fusion at the current moment as the input, the model will perform probability prediction on its belonging category according to the learned knowledge and output one or several most likely corresponding abnormal types, realizing one - key automatic diagnosis of the equipment health status under complex working conditions. In this way, not only is the diagnosis efficiency and accuracy greatly improved, but also a solid technical support is provided for realizing real - time and intelligent operation and maintenance management in the industrial field.

[0048] In summary, the automatic analysis method for the abnormal state of the pressure gauge according to the embodiments of the present application is elucidated. By using the collaborative model based on the autoencoder, it learns the collaborative pattern between the real-time flow rate value sequence and the real-time pressure value sequence under normal working conditions, and infers the expected real-time pressure value sequence from the real-time flow rate value sequence accordingly. By calculating the residual between the actual real-time pressure value sequence and the inferred real-time pressure value sequence, the deviation degree of the collaborative relationship between pressure and flow rate can be effectively quantified. The residual sequence and its temporal characteristics, combined with the pressure temporal characteristics of the target pressure gauge itself, are jointly input into the trained abnormal state diagnosis multi-classifier model, so as to realize accurate and robust diagnosis of the abnormal state of the pressure gauge, and effectively distinguish whether the pressure change is caused by the fault of the pressure gauge itself or the abnormal system process.

[0049] Furthermore, an automatic analysis system for the abnormal state of the pressure gauge is also provided.

[0050] Figure 4 It is a block diagram of the automatic analysis system for the abnormal state of the pressure gauge according to the embodiments of the present application. As Figure 4 shown, the automatic analysis system 300 for the abnormal state of the pressure gauge according to the embodiments of the present application includes: a pressure value acquisition module 310 for acquiring the real-time pressure value sequence of the target pressure gauge; a flow rate value acquisition module 320 for acquiring the real-time flow rate value sequence collected by a flow meter associated with the target pressure gauge; a timestamp alignment module 330 for aligning the timestamps of the real-time pressure value sequence and the real-time flow rate value sequence to obtain the aligned real-time pressure value sequence and the aligned real-time flow rate value sequence; an individual signal feature extraction module 340 for extracting individual signal features from the aligned real-time pressure value sequence to obtain the pressure temporal feature vector of the target pressure gauge; a pressure value inference module 350 for inputting the aligned real-time flow rate value sequence into the collaborative model based on the autoencoder to obtain the inferred real-time pressure value sequence; and an abnormal state diagnosis module 360 for calculating the pressure-flow rate collaborative deviation feature between the inferred real-time pressure value sequence and the real-time pressure value sequence, and inputting the pressure-flow rate collaborative deviation feature and the pressure temporal feature of the target pressure gauge into the trained abnormal state diagnosis multi-classifier model to obtain the abnormal diagnosis result.

[0051] As described above, the automatic analysis system 300 for abnormal pressure gauge states according to the embodiments of the present application can be implemented in various wireless terminals, such as a server with an automatic analysis algorithm for abnormal pressure gauge states. In a possible implementation, the automatic analysis system 300 for abnormal pressure gauge states according to the embodiments of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the automatic analysis system 300 for abnormal pressure gauge states can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the automatic analysis system 300 for abnormal pressure gauge states can also be one of the many hardware modules of the wireless terminal.

[0052] Alternatively, in another example, the automatic analysis system 300 for abnormal pressure gauge states and the wireless terminal can also be separate devices, and the automatic analysis system 300 for abnormal pressure gauge states can be connected to the wireless terminal through a wired and / or wireless network and transmit interaction information in accordance with a predefined data format.

[0053] The embodiments of the present disclosure have been described above. The above description is exemplary and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to technologies in the market, or to enable other ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An automatic analysis method for abnormal states of a pressure gauge, characterized in that, Including: Obtain the real-time pressure value sequence of the target pressure gauge; Obtain the real-time flow rate value sequence collected by the flowmeter associated with the target pressure gauge; Perform timestamp alignment on the real-time pressure value sequence and the real-time flow rate value sequence to obtain the aligned real-time pressure value sequence and the aligned real-time flow rate value sequence; Extract individual signal features from the aligned real-time pressure value sequence to obtain the pressure time series feature vector of the target pressure gauge; Input the aligned real-time flow rate value sequence into the collaborative model based on the autoencoder to obtain the inferred real-time pressure value sequence; Calculate the pressure-flow rate collaborative deviation feature between the inferred real-time pressure value sequence and the real-time pressure value sequence, and input the pressure-flow rate collaborative deviation feature and the pressure time series feature of the target pressure gauge into the trained abnormal state diagnosis multi-classifier model to obtain the abnormal diagnosis result.

2. The automatic analysis method for abnormal states of pressure gauges according to claim 1, characterized in that, Extract individual signal features from the aligned real-time pressure value sequence to obtain the pressure time series feature vector of the target pressure gauge, including: Perform individual signal feature extraction based on LSTM on the aligned real-time pressure value sequence to obtain the pressure time series feature vector of the target pressure gauge.

3. The automatic analysis method for abnormal states of pressure gauges according to claim 1, characterized in that, Calculate the pressure-flow rate collaborative deviation feature between the inferred real-time pressure value sequence and the real-time pressure value sequence, including: Calculate the residual between the inferred real-time pressure value sequence and the real-time pressure value sequence to obtain the real-time pressure value residual sequence; Extract residual time series features from the real-time pressure value residual sequence to obtain the pressure-flow rate collaborative deviation feature vector as the pressure-flow rate collaborative deviation feature.

4. The automatic analysis method for abnormal states of pressure gauges according to claim 1, characterized in that, Input the pressure-flow rate collaborative deviation feature and the pressure time series feature of the target pressure gauge into the trained abnormal state diagnosis multi-classifier model to obtain the abnormal diagnosis result, including: Perform multi-scale joint progressive perception on the pressure-flow rate collaborative deviation feature vector and the pressure time series feature vector of the target pressure gauge to obtain the target pressure gauge main body behavior - collaborative behavior time series joint coding vector; Based on the target pressure gauge main body behavior - collaborative behavior time series joint coding vector, obtain the abnormal diagnosis result.

5. The automatic analysis method for abnormal states of pressure gauges according to claim 4, characterized in that, Perform multi-scale joint progressive perception on the pressure-flow rate collaborative deviation feature vector and the pressure time series feature vector of the target pressure gauge to obtain the target pressure gauge main body behavior - collaborative behavior time series joint coding vector, including: Perform deep non-linear transformation on the pressure-flow rate collaborative deviation feature vector and the pressure time series feature vector of the target pressure gauge respectively to obtain the pressure-flow rate collaborative deviation middle-layer hidden feature coding vector, the target pressure gauge pressure middle-layer time series hidden feature coding vector, the pressure-flow rate collaborative deviation deep-layer hidden feature coding vector, and the target pressure gauge pressure deep-layer time series hidden feature coding vector; Perform multi-level cross-modal time series fusion on the pressure-flow rate collaborative deviation feature vector and the pressure time series feature vector of the target pressure gauge to obtain the target pressure gauge main body behavior - collaborative behavior low-level time series fusion feature coding vector, the target pressure gauge main body behavior - collaborative behavior middle-level time series fusion feature coding vector, and the target pressure gauge main body behavior - collaborative behavior deep-level time series fusion feature coding vector; Perform multi-scale progressive complementary perception fusion on the target pressure gauge main body behavior - collaborative behavior low-level temporal fusion feature encoding vector, the target pressure gauge main body behavior - collaborative behavior middle-level temporal fusion feature encoding vector, and the target pressure gauge main body behavior - collaborative behavior deep-level temporal fusion feature encoding vector to obtain the target pressure gauge main body behavior - collaborative behavior temporal joint encoding vector.

6. The automatic analysis method for abnormal states of pressure gauges according to claim 5, characterized in that Perform deep non-linear transformation on the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure temporal feature vector respectively to obtain the pressure-flow rate collaborative deviation middle-level implicit feature encoding vector, the target pressure gauge pressure middle-level temporal implicit feature encoding vector, the pressure-flow rate collaborative deviation deep-level implicit feature encoding vector, and the target pressure gauge pressure deep-level temporal implicit feature encoding vector, including: Perform multi-level implicit feature extraction on the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure temporal feature vector to obtain the pressure-flow rate collaborative deviation middle-level implicit feature encoding vector, the target pressure gauge pressure middle-level temporal implicit feature encoding vector, the pressure-flow rate collaborative deviation deep-level implicit feature encoding vector, and the target pressure gauge pressure deep-level temporal implicit feature encoding vector.

7. The automatic analysis method for abnormal states of pressure gauges according to claim 6, wherein Perform multi-level cross-modal temporal fusion on the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure temporal feature vector to obtain the target pressure gauge main body behavior - collaborative behavior low-level temporal fusion feature encoding vector, the target pressure gauge main body behavior - collaborative behavior middle-level temporal fusion feature encoding vector, and the target pressure gauge main body behavior - collaborative behavior deep-level temporal fusion feature encoding vector, including: Perform low-level feature fusion on the pressure-flow rate collaborative deviation feature vector and the target pressure gauge pressure temporal feature vector to obtain the target pressure gauge main body behavior - collaborative behavior low-level temporal fusion feature encoding vector; Perform middle-level feature fusion on the pressure-flow rate collaborative deviation middle-level implicit feature encoding vector and the target pressure gauge pressure middle-level temporal implicit feature encoding vector to obtain the target pressure gauge main body behavior - collaborative behavior middle-level temporal fusion feature encoding vector; Perform deep-level feature fusion on the pressure-flow rate collaborative deviation deep-level implicit feature encoding vector and the target pressure gauge pressure deep-level temporal implicit feature encoding vector to obtain the target pressure gauge main body behavior - collaborative behavior deep-level temporal fusion feature encoding vector.

8. The automatic analysis method for abnormal states of pressure gauges according to claim 7, characterized in that, Perform multi-scale progressive complementary perception fusion on the target pressure gauge main body behavior - collaborative behavior low-level temporal fusion feature encoding vector, the target pressure gauge main body behavior - collaborative behavior middle-level temporal fusion feature encoding vector, and the target pressure gauge main body behavior - collaborative behavior deep-level temporal fusion feature encoding vector to obtain the target pressure gauge main body behavior - collaborative behavior temporal joint encoding vector, including: Perform cross-level preliminary interaction fusion based on order parameter phase complementarity on the target pressure gauge main body behavior - collaborative behavior low-level temporal fusion feature encoding vector and the target pressure gauge main body behavior - collaborative behavior middle-level temporal fusion feature encoding vector to obtain the target pressure gauge main body behavior - collaborative behavior cross-level preliminary fusion semantic feature encoding vector; Perform temporal semantic interaction analysis of the target pressure gauge main body behavior - collaborative behavior based on the constraint of the order parameter field and path integral optimization for the cross - level preliminary fusion semantic feature encoding vector and the deep - level temporal fusion feature encoding vector of the target pressure gauge main body behavior - collaborative behavior to obtain the temporal joint encoding vector of the target pressure gauge main body behavior - collaborative behavior.

9. The automatic analysis method for abnormal states of pressure gauges according to claim 8, characterized in that, Based on the temporal joint encoding vector of the target pressure gauge main body behavior - collaborative behavior, obtain the anomaly diagnosis result, including: Input the temporal joint encoding vector of the target pressure gauge main body behavior - collaborative behavior into the trained anomaly state diagnosis multi - classifier model to obtain the anomaly diagnosis result.

10. An automatic analysis system for abnormal states of pressure gauges, characterized in that, Including: A pressure value acquisition module for acquiring the real - time pressure value sequence of the target pressure gauge; A flow velocity value acquisition module for acquiring the real - time flow velocity value sequence collected by a flow meter associated with the target pressure gauge; A timestamp alignment module for aligning the timestamps of the real - time pressure value sequence and the real - time flow velocity value sequence to obtain the aligned real - time pressure value sequence and the aligned real - time flow velocity value sequence; An individual signal feature extraction module for extracting individual signal features from the aligned real - time pressure value sequence to obtain the pressure temporal feature vector of the target pressure gauge; A pressure value inference module for inputting the aligned real - time flow velocity value sequence into a collaborative model based on an auto - encoder to obtain an inferred real - time pressure value sequence; An anomaly state diagnosis module for calculating the pressure - flow velocity collaborative deviation feature between the inferred real - time pressure value sequence and the real - time pressure value sequence, and inputting the pressure - flow velocity collaborative deviation feature and the pressure temporal feature of the target pressure gauge into the trained anomaly state diagnosis multi - classifier model to obtain the anomaly diagnosis result.

Images