Meteorological Instrument Fault Detection System and Method

Abstract

This application relates to the technical field of meteorological instrument detection, and specifically discloses a meteorological instrument fault detection system and method, which includes: first, setting test conditions, collecting response characteristic data of each core component of the meteorological instrument to be tested under the action of the test conditions, and then, using data processing technology based on deep learning to perform significance analysis and aggregation processing on the response characteristic data of each core component to capture the overall response characteristics of the core components of the meteorological instrument to be tested. Furthermore, by performing fine-grained interaction analysis on the test conditions and the overall response characteristics of the core components, the association mode between the test conditions and the response characteristics of the core components is mined, and based on this, it is intelligently identified whether the meteorological instrument to be tested has a fault. In this way, by simulating different working scenarios, the accuracy and timeliness of meteorological instrument fault detection can be effectively improved, thereby ensuring the reliability of meteorological data.

Description

Technical Field

[0001] This application relates to the technical field of meteorological instrument detection, and more specifically, to a meteorological instrument fault detection system and method. Background Art

[0002] As a key device for meteorological observation and prediction, the accuracy and reliability of meteorological instruments are directly related to the accuracy and timeliness of weather forecasts, and have a significant impact on multiple fields such as agricultural production, transportation, environmental protection, and disaster warning. However, during the long-term operation of meteorological instruments, due to various reasons such as environmental factors (such as extreme weather, temperature changes, humidity fluctuations, etc.), mechanical wear, and electronic component aging, faults or performance degradation are inevitable. If the faults cannot be detected and repaired in time, it may lead to distorted observation data, thereby affecting subsequent meteorological analysis and decision-making.

[0003] Traditional methods for detecting faults in meteorological instruments mainly rely on regular manual inspections and maintenance, including visual inspections, functional tests, and simple performance parameter comparisons. Such methods are not only time-consuming and laborious, but also often difficult to capture some hidden or intermittent faults, and cannot effectively identify them at the early stage of fault occurrence, which may lead to the accumulation of data errors and affect the accuracy and timeliness of weather forecasts.

[0004] Therefore, an optimized method and system for detecting faults in meteorological instruments are expected. Summary of the Invention

[0005] This application provides a meteorological instrument fault detection system and method, which can effectively improve the accuracy and timeliness of meteorological instrument fault detection by simulating different working scenarios, thereby ensuring the reliability of meteorological data.

[0006] In a first aspect, a method for detecting faults in meteorological instruments is provided, including:

[0007] Setting test conditions, where the test conditions include input voltage value, input current value, and optical signal intensity value;

[0008] Collecting response characteristic data of each core component of the meteorological instrument under test under the action of the test conditions to obtain a set of core component response characteristic data;

[0009] Performing embedded encoding on the test conditions to obtain a low-dimensional embedded encoding vector of the test conditions;

[0010] Performing feature dynamic aggregation based on embedded encoding on the set of core component response characteristic data to obtain a significantly aggregated encoding vector of core component response characteristics;

[0011] Perform feature fine-grained interaction enhanced by external knowledge on the significant aggregated encoding vector of the response characteristics of the core component and the low-dimensional embedded encoding vector of the test conditions to obtain a fine-grained interaction encoding vector of test conditions-component response characteristics;

[0012] Based on the fine-grained interaction encoding vector of test conditions-component response characteristics, determine whether the meteorological instrument under test has a fault.

[0013] In a second aspect, a meteorological instrument fault detection system is provided, including:

[0014] A test condition setting module for setting test conditions, where the test conditions include input voltage value, input current value, and optical signal intensity value;

[0015] A response characteristic data acquisition module for acquiring the response characteristic data of each core component of the meteorological instrument under test under the action of the test conditions to obtain a set of core component response characteristic data;

[0016] A test condition embedded encoding module for performing embedded encoding on the test conditions to obtain a low-dimensional embedded encoding vector of the test conditions;

[0017] A feature dynamic aggregation processing module for performing feature dynamic aggregation based on embedded encoding on the set of core component response characteristic data to obtain a significant aggregated encoding vector of core component response characteristics;

[0018] A feature fine-grained interaction processing module for performing feature fine-grained interaction enhanced by external knowledge on the significant aggregated encoding vector of the core component response characteristics and the low-dimensional embedded encoding vector of the test conditions to obtain a fine-grained interaction encoding vector of test conditions-component response characteristics;

[0019] A meteorological instrument test result determination module for determining whether the meteorological instrument under test has a fault based on the fine-grained interaction encoding vector of test conditions-component response characteristics.

[0020] A meteorological instrument fault detection system and method provided by this application first set test conditions, collect the response characteristic data of each core component of the meteorological instrument under test under the action of the test conditions, and then use data processing technology based on deep learning to perform significance analysis and aggregation processing on the response characteristic data of each core component to capture the overall response characteristics of the core components of the meteorological instrument under test. Furthermore, by performing fine-grained interaction analysis on the test conditions and the overall response characteristics of the core components, the association pattern between the test conditions and the response characteristics of the core components is mined, and based on this, it is possible to intelligently identify whether the meteorological instrument under test has a fault. In this way, by simulating different working scenarios, the accuracy and timeliness of meteorological instrument fault detection can be effectively improved, thereby ensuring the reliability of meteorological data. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced below. Obviously, the drawings in the following description only relate to some embodiments of the present application and do not limit the present application.

[0022] Figure 1 It is a schematic flowchart of the method for detecting faults of meteorological instruments according to the embodiments of the present application.

[0023] Figure 2 It is a schematic diagram of data flow of the method for detecting faults of meteorological instruments according to the embodiments of the present application.

[0024] Figure 3 It is a schematic flowchart of step S4 in the method for detecting faults of meteorological instruments according to the embodiments of the present application.

[0025] Figure 4 It is a schematic flowchart of step S5 in the method for detecting faults of meteorological instruments according to the embodiments of the present application.

[0026] Figure 5 It is a schematic block diagram of the system for detecting faults of meteorological instruments according to the embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts also belong to the scope of protection of the present application.

[0028] In view of the above technical problems, the technical concept of the present application is as follows: First, set test conditions, collect the response characteristic data of each core component of the meteorological instrument to be tested under the action of the test conditions, and then, use the data processing technology based on deep learning to perform significance analysis and aggregation processing on the response characteristic data of each core component to capture the overall response characteristics of the core components of the meteorological instrument to be tested. Furthermore, by performing fine-grained interaction analysis on the test conditions and the overall response characteristics of the core components, the association pattern between the test conditions and the response characteristics of the core components is mined, and based on this, it is possible to intelligently identify whether the meteorological instrument to be tested has a fault. In this way, by simulating different working scenarios, the accuracy and timeliness of meteorological instrument fault detection can be effectively improved, thereby ensuring the reliability of meteorological data.

[0029] Such as Figure 1 and Figure 2As shown, the method for detecting faults in a meteorological instrument includes: S1, setting test conditions, where the test conditions include an input voltage value, an input current value, and an optical signal intensity value; S2, collecting response characteristic data of each core component of the meteorological instrument under test under the action of the test conditions to obtain a set of core component response characteristic data; S3, performing embedded coding on the test conditions to obtain a low-dimensional embedded coding vector of the test conditions; S4, performing feature dynamic aggregation based on embedded coding on the set of core component response characteristic data to obtain a significantly aggregated coding vector of core component response characteristics; S5, performing feature fine-grained interaction based on external knowledge enhancement on the significantly aggregated coding vector of core component response characteristics and the low-dimensional embedded coding vector of the test conditions to obtain a fine-grained interaction coding vector of test condition-component response characteristics; S6, determining whether the meteorological instrument under test has a fault based on the fine-grained interaction coding vector of test condition-component response characteristics.

[0030] Exemplarily, in step S1, test conditions are set, and the test conditions include an input voltage value, an input current value, and an optical signal intensity value. It should be understood that the input voltage and current values directly affect the working efficiency and stability of electronic components, while the optical signal intensity has an important impact on components that rely on optical sensing in meteorological instruments. By setting the test conditions, the actual working scenario of the meteorological instrument can be simulated to obtain comprehensive information about each core component of the device under test, which helps to discover potential problems or early signs of faults.

[0031] Exemplarily, in step S2, response characteristic data of each core component of the meteorological instrument under test are collected under the action of the test conditions to obtain a set of core component response characteristic data. It should be understood that each core component in the meteorological instrument may have different response characteristics to input conditions (such as voltage, current, optical signal intensity). Therefore, by collecting the response characteristic data of all core components, the performance of each component under specific test conditions can be comprehensively understood, and a comprehensive view of the overall health status of the meteorological instrument under test can be obtained, providing a comprehensive data basis for subsequent fault detection and analysis. In the embodiments of the present application, the response characteristic data includes but is not limited to voltage, current, power, component surface temperature, light intensity, etc.

[0032] Exemplarily, in step S3, the test conditions are embedded and encoded to obtain a low-dimensional embedded encoding vector of the test conditions. It should be understood that considering that the response characteristic data of the core component is a comprehensive reflection of the meteorological instrument under the combined action of the input voltage value, input current value, and optical signal intensity value, therefore, in order to capture the internal relationship between the multi-dimensional information in the test conditions, the present application further performs embedded encoding on the test conditions to compress the multi-dimensional information of the test conditions into a unified low-dimensional feature space, so as to facilitate the identification and expression of the potential correlation between the multi-dimensional information and realize the combined analysis of the multi-dimensional test conditions. In the embodiment of the present application, a fully connected neural network model is used to train and generate a test condition embedded encoding matrix, and this is used to perform embedded encoding on the test conditions to obtain a low-dimensional embedded encoding vector of the test conditions.

[0033] Exemplarily, in step S4, a feature dynamic aggregation based on embedded encoding is performed on the set of the core component response characteristic data to obtain a significantly aggregated encoding vector of the core component response characteristics. It should be understood that considering that there are many core components in the meteorological instrument and there is a certain non-linear correlation relationship between the response characteristic data of each core component, therefore, in order to capture the interaction and associated influence between each core component, and further to discover the overall significant response pattern of the core components of the meteorological instrument, the present application further performs feature dynamics analysis and significant aggregation processing on the set of the core component response characteristic data to reveal the overall response characteristics of the core components of the meteorological instrument.

[0034] In one embodiment, as Figure 3 shown, performing a feature dynamic aggregation based on embedded encoding on the set of the core component response characteristic data to obtain a significantly aggregated encoding vector of the core component response characteristics includes: S41, respectively passing each core component response characteristic data in the set of the core component response characteristic data through an embedding layer to obtain a set of semantic embedded encoding vectors of the core component response characteristics; S42, calculating a pseudo-anchored aggregation center representation vector of the core component response characteristics based on the static potential energy distribution of the features of the set of the semantic embedded encoding vectors of the core component response characteristics; S43, calculating the aggregation movement direction of each semantic embedded encoding vector of the core component response characteristics in the set of the semantic embedded encoding vectors of the core component response characteristics relative to the pseudo-anchored aggregation center representation vector of the core component response characteristics to obtain a set of aggregation movement directions of the core component response characteristics; S44, based on the set of the aggregation movement directions of the core component response characteristics, dynamically aggregating the set of the semantic embedded encoding vectors of the core component response characteristics towards the pseudo-anchored aggregation center representation vector of the core component response characteristics to obtain the significantly aggregated encoding vector of the core component response characteristics.

[0035] Exemplarily, in step S41, each of the core component response characteristic data in the set of core component response characteristic data is respectively passed through an embedding layer to obtain a set of core component response characteristic semantic embedding encoding vectors. It should be understood that each core component response characteristic data is respectively embedded and encoded through the embedding encoding technology to map it to a unified feature space, obtaining the semantic embedding representation of each core component response characteristic data, thereby obtaining a set of core component response characteristic semantic embedding encoding vectors.

[0036] In one embodiment, in step S42, based on the characteristic static potential energy distribution of the set of core component response characteristic semantic embedding encoding vectors, a core component response characteristic pseudo-anchoring aggregation center representation vector is calculated, including: respectively inputting each of the core component response characteristic semantic embedding encoding vectors in the set of core component response characteristic semantic embedding encoding vectors into a characteristic static potential energy measurement network to obtain a set of core component response characteristic static potential energy measurement coefficients; inputting the set of core component response characteristic static potential energy measurement coefficients into a characteristic energy level screening gating unit to obtain a set of core component response characteristic static potential energy weight factors; based on the set of core component response characteristic static potential energy weight factors, calculating the weighted sum of the set of core component response characteristic semantic embedding encoding vectors to obtain the core component response characteristic pseudo-anchoring aggregation center representation vector. Specifically, this process can be expressed by the formula as follows:

[0037]

[0038]

[0039]

[0040]

[0041]

[0042] where represents the set of core component response characteristic semantic embedding encoding vectors, , , and respectively represent the first, second, th, and th core component response characteristic semantic embedding encoding vectors in the set of core component response characteristic semantic embedding encoding vectors, is the number of feature vectors in the set of core component response characteristic semantic embedding encoding vectors, represents the th core component response characteristic semantic embedding encoding vector, and The eigenvalue of a position, and respectively represent the mean value and variance of the eigenvalues of the semantic embedding coding vectors of the response characteristics of the th core component, is the eigenvalue scale value of the semantic embedding coding vector of the response characteristics of the core component, is the static potential energy metric coefficient of the semantic embedding coding vector of the response characteristics of the th core component, represents the th normalized static potential energy metric coefficient of the response characteristics of the core component, represents the gating mask function, represents the gating threshold, represents the th static potential energy weight factor of the response characteristics of the core component, represents the pseudo-anchored aggregation center representation vector of the response characteristics of the core component.

[0043] That is, by measuring the characteristic potential energy of the semantic embedding coding vectors of the response characteristics of each core component, the potential energy level of the response characteristic data of each core component under static conditions, as well as its importance and influence under the global response characteristic distribution of the core component, are revealed, thereby obtaining a set of static potential energy metric coefficients of the response characteristics of the core component. Then, through the gating mechanism, the set of static potential energy metric coefficients of the response characteristics of the core component obtained is subjected to gating mask processing to screen out individual characteristics that contribute more to the overall response characteristics of the core components of the meteorological instrument and assign corresponding weight factors to them, while suppressing individual characteristics with less contribution to reduce the impact of noise and irrelevant information on the final fault detection. Furthermore, based on the obtained weight factors, the set of semantic embedding coding vectors of the response characteristics of the core component is weighted and aggregated to reveal the trend center of the response characteristics of each core component, generating a pseudo-anchored aggregation center representation vector of the response characteristics of the core component, so as to reflect the essential response characteristics of the core components of the meteorological instrument, thereby providing effective guidance for the subsequent aggregation process of the response characteristics of individual components.

[0044] Specifically, first, the feature static potential energy metric network is used to evaluate the stability or "potential energy" of the semantic embedding encoding vectors of the response characteristics of each core component under the current test conditions. This step helps identify which component behavior patterns are stable and which are unstable under given conditions. Then, through the feature energy level screening gating unit, the weights of each response characteristic are further filtered and adjusted according to the metric results. The role of the gating unit is to strengthen the features of the core components that have a greater impact on the overall response characteristics of the meteorological instrument, while weakening or ignoring the features that contribute less to the global behavior or may be outliers. Such processing helps remove irrelevant information and makes subsequent analysis more focused on meaningful data. Finally, a pseudo-anchoring aggregation center representation vector is determined based on the calculated set of static potential energy weight factors of the core component response characteristics. This aggregation center actually reflects the average response trend of all measured core components under specific test conditions, and it is used as a reference point to measure the position of other individual components relative to the overall performance. In this way, the comprehensive response characteristics of the entire meteorological instrument in different working scenarios can be captured more accurately, providing a solid foundation for subsequent fine-grained interaction analysis. In addition, this method also improves the system's sensitivity to potential failures because it emphasizes those important changes that may indicate a problem is about to occur, even if these changes initially seem insignificant.

[0045] That is, each core component response characteristic semantic embedding encoding vector in the set of core component response characteristic semantic embedding encoding vectors is input into the feature static potential energy metric network respectively to obtain a set of core component response characteristic static potential energy metric coefficients, including: performing Z-score standardization processing on the core component response characteristic semantic embedding encoding vector to obtain a standardized core component response characteristic semantic embedding encoding vector; calculating the sum of the cubes of each eigenvalue in the standardized core component response characteristic semantic embedding encoding vector divided by the feature scale value of the standardized core component response characteristic semantic embedding encoding vector to obtain the core component response characteristic static potential energy metric coefficient.

[0046] Exemplarily, in step S43, the aggregation movement direction of each core component response characteristic semantic embedding encoding vector in the set of core component response characteristic semantic embedding encoding vectors relative to the core component response characteristic pseudo-anchoring aggregation center representation vector is calculated to obtain a set of core component response characteristic aggregation movement directions. In one embodiment, the core component response characteristic aggregation movement direction is the arccosine function value between the core component response characteristic semantic embedding encoding vector and the core component response characteristic pseudo-anchoring aggregation center representation vector. Specifically, this process can be expressed by the formula as follows:

[0047]

[0048] Among them, represents the -th core component response characteristic semantic embedding coding vector in the set of core component response characteristic semantic embedding coding vectors, represents the core component response characteristic pseudo-anchoring aggregation center representation vector, represents the norm of the vector, is the arccosine function, represents the aggregation movement direction of the -th core component response characteristic semantic embedding coding vector relative to the core component response characteristic pseudo-anchoring aggregation center representation vector.

[0049] Specifically, when the response characteristics of all core components are converted into low-dimensional semantic embedding coding vectors, these vectors together form a spatial distribution reflecting the overall health status of the device. By determining a pseudo-anchoring aggregation center representation vector, that is, an average or typical state of all core component response characteristics, it can be used as a reference point to evaluate whether the behavior of a single component deviates from the expectation. Then, for each core component response characteristic semantic embedding coding vector, calculate the angular difference between it and the pseudo-anchoring aggregation center. Usually, the arccosine function is used to measure the included angle between these two vectors. This angular difference actually represents the "movement direction" of the core component relative to the overall response characteristics, revealing its dynamic role in the global response pattern. This aggregation movement direction not only describes how individual components change with test conditions, but also shows their interrelationships and synergistic effects in the entire meteorological instrument. For example, in the normal operating state, most core components may show similar movement directions, indicating that they respond to environmental changes in a coordinated manner; in case of a failure, some abnormal components may show significantly different movement directions, suggesting that their behavior has deviated from the normal group pattern. Therefore, by analyzing these aggregation movement directions, potential problems or early failure signs can be identified earlier, even if these problems have not yet led to obvious performance degradation.

[0050] That is to say, by calculating the aggregation movement direction of each core component response characteristic semantic embedding coding vector relative to the core component response characteristic pseudo-anchoring aggregation center representation vector, the change trend of each individual feature relative to the center point is quantified, revealing the dynamic relationship and evolution path of each individual feature within the core component response characteristic space, so as to facilitate the identification of the relative relationship between features and the group behavior pattern.

[0051] Exemplarily, in step S44, based on the set of aggregation movement directions of the core component response characteristics, the set of semantic embedding encoding vectors of the core component response characteristics is dynamically aggregated towards the core component response characteristics pseudo-anchoring aggregation center representation vector to obtain the core component response characteristics significant aggregation encoding vector. Specifically, this process can be expressed by the formula:

[0052]

[0053] wherein, represents the -th semantic embedding encoding vector of the core component response characteristics in the set of semantic embedding encoding vectors of the core component response characteristics, is the static potential energy metric coefficient of the -th semantic embedding encoding vector of the core component response characteristics, represents the core component response characteristics pseudo-anchoring aggregation center representation vector, represents the aggregation movement direction of the -th semantic embedding encoding vector of the core component response characteristics relative to the core component response characteristics pseudo-anchoring aggregation center representation vector, represents the sine function, and represent the weight matrix and the bias term respectively, represents the core component response characteristics significant aggregation encoding vector.

[0054] It should be understood that by calculating the aggregation movement direction of each semantic embedding encoding vector of the core component response characteristics relative to the pseudo-anchoring aggregation center, the present application has obtained an understanding of how each component changes relative to the global behavior under specific test conditions. Next, using this aggregation movement direction information, the semantic embedding encoding vectors of each core component response characteristics can be guided to dynamically adjust towards the pseudo-anchoring aggregation center. This adjustment is not a simple averaging process, but a weighted aggregation according to the specific movement direction of each component and its importance in the overall response. This can ensure that the component features that contribute more to the overall response characteristics are fully emphasized, while the relatively less important or possibly noisy data are appropriately suppressed. During the dynamic aggregation process, the semantic embedding encoding vectors of the core component response characteristics will gradually approach the pseudo-anchoring aggregation center representation vector, but at the same time retain their unique "movement trajectories". This means that the finally formed significant aggregation encoding vector not only represents a static average state, but also integrates the dynamic response characteristics of all core components under different conditions. Such an encoding vector can more accurately reflect the overall health status of the meteorological instrument and its performance trends in various working scenarios.

[0055] That is, based on the aggregated motion direction and the characteristic static potential energy metric coefficient of the semantic embedding coding vectors of the response characteristics of each core component, the semantic embedding coding vectors of the response characteristics of each core component are dynamically aggregated towards the pseudo-anchoring aggregation center representation vector of the response characteristics of the core component, so as to realize the dynamic characterization of the overall response characteristics of the core components of the meteorological instrument, and obtain the significantly aggregated coding vectors of the response characteristics of the core components.

[0056] Exemplarily, in step S5, a feature fine-grained interaction based on external knowledge enhancement is performed on the significantly aggregated coding vectors of the response characteristics of the core components and the low-dimensional embedding coding vectors of the test conditions to obtain the fine-grained interaction coding vectors of the test conditions-component response characteristics. It should be understood that considering that when the meteorological instrument is in a normal working state, its core components will exhibit specific response patterns for specific test conditions, and when the core components fail, this response pattern will change. Therefore, the present application further performs interaction analysis on the significantly aggregated coding vectors of the response characteristics of the core components and the low-dimensional embedding coding vectors of the test conditions to reveal the association pattern between the response data of the core components of the meteorological instrument and the current test conditions, so as to realize the detection of potential faults.

[0057] In one embodiment, as Figure 4 shown, performing a feature fine-grained interaction based on external knowledge enhancement on the significantly aggregated coding vectors of the response characteristics of the core components and the low-dimensional embedding coding vectors of the test conditions to obtain the fine-grained interaction coding vectors of the test conditions-component response characteristics includes: S51, based on external knowledge, performing feature interaction optimization based on the attention mechanism on the significantly aggregated coding vectors of the response characteristics of the core components and the low-dimensional embedding coding vectors of the test conditions to obtain the fine-grained interaction feature matrix of the external knowledge-optimized test conditions-component response characteristics; S52, based on the fine-grained interaction feature matrix of the external knowledge-optimized test conditions-component response characteristics, performing feature modulation optimization on the significantly aggregated coding vectors of the response characteristics of the core components and the low-dimensional embedding coding vectors of the test conditions respectively to obtain the optimized significantly aggregated coding vectors of the response characteristics of the core components and the optimized low-dimensional embedding coding vectors of the test conditions; S53, calculating the division of the optimized significantly aggregated coding vectors of the response characteristics of the core components and the optimized low-dimensional embedding coding vectors of the test conditions at each position point to obtain the fine-grained interaction coding vectors of the test conditions-component response characteristics.

[0058] In one embodiment, in step S51, based on external knowledge, feature interaction optimization based on the attention mechanism is performed on the significant aggregation encoding vector of the core component response characteristics and the low-dimensional embedding encoding vector of the test conditions to obtain an external knowledge-optimized test condition-component response characteristic fine-grained interaction feature matrix, including: inputting the significant aggregation encoding vector of the core component response characteristics and the low-dimensional embedding encoding vector of the test conditions into a fine-grained feature interaction network to obtain a test condition-component response characteristic fine-grained interaction feature matrix; inputting the test condition-component response characteristic fine-grained interaction feature matrix into an attention unit based on external knowledge to obtain the external knowledge-optimized test condition-component response characteristic fine-grained interaction feature matrix. Specifically, this process can be expressed by the formula:

[0059]

[0060]

[0061] Wherein, represents the significant aggregation encoding vector of the core component response characteristics, represents the low-dimensional embedding encoding vector of the test conditions, represents the transpose of the vector, represents matrix multiplication operation, represents the test condition-component response characteristic fine-grained interaction feature matrix, and represent the learnable memory parameter matrices of the attention unit based on external knowledge, represents the normalization function, represents the external knowledge-optimized test condition-component response characteristic fine-grained interaction feature matrix.

[0062] That is, by performing fine-grained feature interaction on the significant aggregation encoding vector of the core component response characteristics and the low-dimensional embedding encoding vector of the test conditions, the semantic association between the two is understood, and a test condition-component response characteristic fine-grained interaction feature matrix is generated. Subsequently, the generated test condition-component response characteristic fine-grained interaction feature matrix is fed into an attention unit based on external knowledge to further optimize the test condition-component response characteristic fine-grained interaction feature matrix using external knowledge, thereby enhancing the model's understanding ability of the working mode of meteorological instruments. In a specific implementation, external knowledge such as the design principle, working principle of meteorological instruments, and the operating environment information of the instruments is used to train the parameter matrix of the attention unit, guiding the attention unit to perform weighted optimization on the test condition-component response characteristic fine-grained interaction feature matrix, enabling it to identify and strengthen the important response patterns of meteorological instruments under the current test conditions, so as to capture the potential connection between the test conditions and the core component response data within a larger range.

[0063] In one embodiment, in step S52, based on the external knowledge, the fine-grained interaction feature matrix of the test condition-component response characteristics is optimized, and the significant aggregation coding vector of the core component response characteristics and the low-dimensional embedding coding vector of the test condition are respectively subjected to feature modulation optimization to obtain the optimized significant aggregation coding vector of the core component response characteristics and the optimized low-dimensional embedding coding vector of the test condition, including: performing a linear transformation on the significant aggregation coding vector of the core component response characteristics to obtain a first query feature vector and a first value feature vector, and using the fine-grained interaction feature matrix of the external knowledge-optimized test condition-component response characteristics as a key matrix, and inputting the first query feature vector, the first value feature vector, and the key matrix into the fine-grained modulation module based on the Transformer structure to obtain the optimized significant aggregation coding vector of the core component response characteristics; performing a linear transformation on the low-dimensional embedding coding vector of the test condition to obtain a second query feature vector and a second value feature vector, and using the fine-grained interaction feature matrix of the external knowledge-optimized test condition-component response characteristics as a key matrix, and inputting the second query feature vector, the second value feature vector, and the key matrix into the fine-grained modulation module based on the Transformer structure to obtain the optimized low-dimensional embedding coding vector of the test condition. Specifically, this process can be represented by the formula:

[0064]

[0065] Wherein, and respectively represent the first query embedding matrix and the first value embedding matrix, and respectively represent the first query feature vector and the first value feature vector, and respectively represent the second query embedding matrix and the second value embedding matrix, and respectively represent the second query feature vector and the second value feature vector, , , and respectively represent different bias terms, represents the normalized exponential function, represents the feature scale value of the fine-grained feature interaction matrix optimized by the external knowledge, and respectively represent the optimized significant aggregation coding vector of the core component response characteristics and the optimized low-dimensional embedding coding vector of the test condition.

[0066] Specifically, by performing a linear transformation on the significantly aggregated encoding vectors of the core component response characteristics to obtain the first query feature vector and the first value feature vector, and using the test condition-component response characteristic fine-grained interaction feature matrix optimized with external knowledge as the key matrix, a deeper exploration of the interactions between core components can be achieved. This key matrix contains key information learned from a large amount of historical data, which reflects how each component works together or changes independently under different test conditions. When the present application inputs the first query feature vector, the first value feature vector, and the key matrix into the fine-grained modulation module based on the Transformer structure, this module can weightedly aggregate relevant information according to the similarity between the query vector (i.e., the current state of the core component) and the elements in the key matrix, thereby generating a more accurate and representative optimized significantly aggregated encoding vector of the core component response characteristics. This step enhances the model's ability to capture complex relationships between core components, enabling accurate identification of abnormal behaviors even in the face of non-linear or multi-variable interaction effects.

[0067] Similarly, a similar process is also carried out for the low-dimensional embedded encoding vector of the test condition: the second query feature vector and the second value feature vector are obtained through linear transformation, and the same key matrix is used again for fine-grained modulation. The goal here is to optimize the test condition representation to better reflect the overall response characteristics of the meteorological instrument under specific conditions. In this way, not only can the factors crucial for a given test condition be strengthened, but also irrelevant information can be suppressed, reducing noise interference. The finally obtained optimized low-dimensional embedded encoding vector of the test condition provides a cleaner and more insightful data basis for subsequent analysis.

[0068] Throughout the process, the fine-grained modulation module based on the Transformer structure plays a key role. This architecture is good at handling long-sequence dependency problems and can effectively model complex context relationships. By introducing the self-attention mechanism, it can flexibly adjust the importance of features at different positions, ensuring that each core component or test condition can be fully considered from a global perspective. In addition, due to the use of the test condition-component response characteristic fine-grained interaction feature matrix optimized with external knowledge as guidance, the learning efficiency and generalization ability of the model are further improved.

[0069] That is, the fine-grained interaction feature matrix of test condition-component response characteristics optimized with external knowledge is used as the key matrix. At the same time, a first query feature vector and a first value feature vector are constructed based on the significantly aggregated encoding vector of the core component response characteristics, and a second query feature vector and a second value feature vector are constructed based on the low-dimensional embedded encoding vector of the test condition. The self-attention mechanism of the Transformer structure is used to realize the information exchange and integration between internal features and external knowledge, so as to ensure that the significantly aggregated encoding vector of the core component response characteristics and the low-dimensional embedded encoding vector of the test condition can benefit from external knowledge to improve the quality of their feature representation.

[0070] Exemplarily, in step S53, the fine-grained interaction encoding vector of test condition-component response characteristics is obtained by dividing the significantly aggregated encoding vector of the optimized core component response characteristics and the low-dimensional embedded encoding vector of the optimized test condition at each position point.

[0071]

[0072] Wherein, and respectively represent the significantly aggregated encoding vector of the optimized core component response characteristics and the low-dimensional embedded encoding vector of the optimized test condition, represents the fine-grained interaction encoding vector of test condition-component response characteristics.

[0073] Specifically, when a core component exhibits anomalies under given test conditions, the corresponding optimized response characteristic encoding vector may show significant fluctuations or proportions deviating from the normal range in some dimensions. Through element-wise division by position, this application can identify the specific locations of these abnormal changes and evaluate their relative importance under the entire test conditions. For example, if the readings of a certain sensor suddenly increase at a specific voltage level while remaining stable under other conditions, the result after the division operation will highlight this point, helping this application quickly lock down the problem. In addition, this method also helps to eliminate the interference between feature values of different scales. Since the original data may come from various types and magnitudes of measurements (such as voltage, current, temperature, etc.), direct comparison may cause some high-magnitude features to obscure low-magnitude but equally important information. Through the division operation, this application shifts the focus to the relative changes between features, making even very subtle differences clearly visible. This is particularly important for detecting small deviations that are not easily noticeable but indicate potential failures. The finally generated test condition-component response characteristic fine-grained interaction encoding vector integrates the information of the optimized core component response characteristics and the test conditions, forming a representation that comprehensively reflects their interaction. Such an encoding vector not only contains knowledge about the overall performance of the meteorological instrument under the current working state but also retains the unique response characteristics of individual components, providing rich context information for subsequent analysis. It can help the model more accurately identify which combination patterns are normal and which may be signs of failures, thereby supporting earlier and more reliable fault warnings.

[0074] That is, through element-wise division by position, the optimized core component response characteristic significant aggregation encoding vector and the optimized test condition low-dimensional embedding encoding vector are subjected to per-position interaction response encoding to obtain the test condition-component response characteristic fine-grained interaction encoding vector, so as to comprehensively reflect the performance of the meteorological instrument under the current test conditions, capture the association pattern between the core component response data of the meteorological instrument and the current test conditions, and thus provide a more intelligent and accurate judgment basis for subsequent fault detection.

[0075] Exemplarily, in step S6, based on the test condition-component response characteristic fine-grained interaction coding vector, it is determined whether the meteorological instrument under test has a fault. In one embodiment, determining whether the meteorological instrument under test has a fault based on the test condition-component response characteristic fine-grained interaction coding vector includes: inputting the test condition-component response characteristic fine-grained interaction coding vector into a classifier-based fault detection module to obtain a fault detection result, and the fault detection result is used to indicate whether the meteorological instrument under test has a fault. It should be understood that the test condition-component response characteristic fine-grained interaction coding vector already contains rich information about the response characteristics of the core components of the meteorological instrument under different working conditions. However, although this information is detailed, it is still not intuitive or clear enough to directly judge the device state. Therefore, the purpose of introducing the classifier is to analyze these coding vectors through machine learning algorithms, so as to achieve the conversion from complex data to a simple binary decision (normal / fault). The classifier can automatically process a large number of complex coding vectors and output a clear result according to predefined criteria - that is, whether the meteorological instrument is in a fault state. This not only greatly reduces the need for manual intervention and improves the detection efficiency, but also through training, the classifier can learn to identify which combination patterns are usually associated with normal operation, while which ones indicate potential problems. This pattern recognition ability enables it to detect anomalies at an early stage, even if these anomalies have not yet led to obvious performance degradation. Further, in a specific embodiment, the classifier-based fault detection module uses the Random Forest algorithm.

[0076] Here, in the case where the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedding coding vector respectively represent the semantic aggregation coding feature of the response characteristic data of the core component under the predetermined test condition and the low-dimensional embedding feature of the predicted test condition, when performing feature fine-grained interaction based on external knowledge modulation, considering that the feature order levels of the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedding coding vector will cause semantic space mismatch when the external knowledge acts on the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedding coding vector respectively, so that the test condition-component response characteristic fine-grained interaction coding vector has a fused distribution space structure difference, affecting the classification regression convergence consistency, and thus affecting the accuracy of the fault detection result obtained by the classifier-based fault detection module.

[0077] Preferably, obtaining the fault detection result by passing the test condition-component response characteristic fine-grained interaction coding vector through the classifier-based fault detection module includes:

[0078] Determine the mean eigenvalue corresponding to the fine-grained interaction coding vector of the test condition-component response characteristic , and after performing a dot addition of the fine-grained interaction coding vector of the test condition-component response characteristic and the mean eigenvalue, then perform a dot multiplication of the dot addition result and the mean eigenvalue to obtain the fine-grained interaction position polysemy interaction vector of the test condition-component response characteristic , where represents the fine-grained interaction coding vector of the test condition-component response characteristic represents the mean eigenvalue corresponding to the fine-grained interaction coding vector of the test condition-component response characteristic, that is, calculate the mean of all eigenvalues of the vector represents dot addition represents dot multiplication;

[0079] Determine the standard deviation of the eigenvalues corresponding to the fine-grained interaction coding vector of the test condition-component response characteristic , and further perform a dot subtraction of the dot addition vector of the fine-grained interaction position polysemy interaction vector of the test condition-component response characteristic and the standard deviation of the eigenvalues from the fine-grained interaction coding vector of the test condition-component response characteristic to obtain the fine-grained interaction position balance metric vector of the test condition-component response characteristic , where represents the standard deviation of the eigenvalues corresponding to the fine-grained interaction coding vector of the test condition-component response characteristic, that is, calculate the standard deviation of all eigenvalues of the vector represents dot addition represents dot subtraction;

[0080] Take the reciprocal of each bit of the fine-grained interaction position balance metric vector of the test condition-component response characteristic , perform a dot addition with the fine-grained interaction position polysemy interaction vector of the test condition-component response characteristic, and then perform a dot multiplication with the fine-grained interaction coding vector of the test condition-component response characteristic to obtain the fine-grained interaction position inference bias vector of the test condition-component response characteristic ;

[0081] Multiply the sum of the mean eigenvalue and the standard deviation of the eigenvalues by the scaling weight hyperparameter , and then perform a dot multiplication with the fine-grained interaction position inference bias vector of the test condition-component response characteristic to obtain the optimized fine-grained interaction coding vector of the test condition-component response characteristic .

[0082] Based on this, due to the possible lack of spatial structure in the feature set of the fine-grained interaction coding vector of the test condition-component response characteristics in the high-dimensional space, which leads to inconsistent convergence of the implicit inference of spatial structure information based on features in image semantic classification regression, the global characteristic category of the fine-grained interaction coding vector of the test condition-component response characteristics is utilized and a description is constructed with reference to its position to establish a far-field characteristic association, thereby establishing the characteristic proximity correlation of the fine-grained interaction coding vector of the test condition-component response characteristics. Furthermore, by estimating the unconstrained characteristic value points of the fine-grained interaction coding vector of the test condition-component response characteristics, the position polysemy information of the object characteristic value is captured, and then the position inference bias recognition ability of the characteristic set cluster of the fine-grained interaction coding vector of the test condition-component response characteristics is enhanced. In this way, the convergence consistency of classification regression is improved, and the accuracy of the fault detection result obtained by the fine-grained interaction coding vector of the test condition-component response characteristics through the classifier-based fault detection module is improved.

[0083] In summary, the meteorological instrument fault detection method according to the embodiments of the present application is elucidated. First, test conditions are set, and the response characteristic data of each core component of the tested meteorological instrument are collected under the action of the test conditions. Then, significant analysis and aggregation processing are performed on the response characteristic data of each core component by using a deep learning-based data processing technology to capture the overall response characteristics of the core components of the tested meteorological instrument. Furthermore, by performing fine-grained interaction analysis on the test conditions and the overall response characteristics of the core components, the association pattern between the test conditions and the response characteristics of the core components is mined, and based on this, it is intelligently identified whether the tested meteorological instrument has a fault. In this way, by simulating different working scenarios, the accuracy and timeliness of meteorological instrument fault detection can be effectively improved, thereby ensuring the reliability of meteorological data.

[0084] Figure 5 It is a schematic block diagram of the meteorological instrument fault detection system according to the embodiments of the present application. As Figure 5As shown, the meteorological instrument fault detection system 100 includes: a test condition setting module 110 for setting test conditions, where the test conditions include input voltage value, input current value, and optical signal intensity value; a response characteristic data acquisition module 120 for acquiring response characteristic data of each core component of the meteorological instrument under test under the action of the test conditions to obtain a set of core component response characteristic data; a test condition embedding encoding module 130 for performing embedding encoding on the test conditions to obtain a low-dimensional embedding encoding vector of the test conditions; a feature dynamic aggregation processing module 140 for performing feature dynamic aggregation based on the embedding encoding on the set of core component response characteristic data to obtain a significantly aggregated encoding vector of the core component response characteristics; a feature fine-grained interaction processing module 150 for performing feature fine-grained interaction based on external knowledge enhancement on the significantly aggregated encoding vector of the core component response characteristics and the low-dimensional embedding encoding vector of the test conditions to obtain a fine-grained interaction encoding vector of the test condition-component response characteristics; and a meteorological instrument test result determination module 160 for determining whether the meteorological instrument under test has a fault based on the fine-grained interaction encoding vector of the test condition-component response characteristics.

[0085] Here, those skilled in the art can understand that the specific operations of each module and unit in the above meteorological instrument fault detection system have been introduced in detail in the description of the meteorological instrument fault detection method above with reference to Figures 1 to 4 and thus, the repeated description thereof will be omitted.

[0086] The embodiments of the present application also provide a computer program product, which includes computer program code. When the computer program code runs on a computer, it enables the computer to implement the methods in the above embodiments of the present application.

[0087] The embodiments of the present application also provide a computer-readable storage medium, which stores computer instructions. When the computer instructions run on a computer, it enables the computer to implement the methods in the above embodiments of the present application.

[0088] The embodiments of the present application also provide a chip, including a circuit for executing the methods in the above embodiments of the present application.

[0089] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0090] In the description of the embodiments of the present application, unless otherwise specified, " / " means "or". For example, A / B may mean A or B; herein, "and / or" is an association relationship describing associated objects, indicating that there can be three relationships. For example, A and / or B may mean: A exists alone, A and B exist simultaneously, and B exists alone. In the present application, "at least one" means one or more, and "a plurality of" means two or more. "At least one (item)" or its similar expression refers to any combination of these items, including any combination of single items or plural items. For example, at least one (item) of a, b, or c may mean: a, b, c, a - b, a - c, b - c, or a - b - c, where a, b, and c may be single or multiple.

[0091] In the embodiments of the present application, prefix words such as "first" and "second" are only used to distinguish different described objects, and have no restrictive effect on the position, order, priority, quantity, content, etc. of the described objects. In the embodiments of the present application, the use of ordinal numbers and other prefix words for distinguishing described objects does not constitute a limitation on the described objects. For the statement of the described objects, refer to the description in the claims or the context of the embodiments, and no redundant limitation should be formed due to the use of such prefix words.

[0092] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.

[0093] In each embodiment of the present application, if there is no special explanation and logical conflict, the terms and / or descriptions between the embodiments are consistent and can be mutually referred to. The technical features in different embodiments can be combined to form new embodiments according to their internal logical relationships.

[0094] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0095] In addition, in each embodiment of the present application, each functional unit may be integrated into one processing unit, may exist physically alone for each unit, or two or more units may be integrated into one unit.

[0096] As described above, the above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A meteorological instrument fault detection method, characterized in that: include: Setting test conditions, wherein the test conditions include an input voltage value, an input current value, and an optical signal strength value; Collecting response characteristic data of each core component of the tested meteorological instrument under the test conditions to obtain a set of core component response characteristic data; Embedding the test condition to obtain a low-dimensional embedded coding vector of the test condition; Performing dynamic feature aggregation based on embedded coding on the set of core component response characteristic data to obtain a core component response characteristic significant aggregation coding vector; Performing feature fine-grained interaction based on external knowledge enhancement on the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedded coding vector to obtain a test condition-component response characteristic fine-grained interaction coding vector; Determining whether a fault occurs in the meteorological instrument under test based on the test condition-component response characteristic fine-grained interactive coding vector; Performing feature fine-grained interaction based on external knowledge enhancement on the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedded coding vector to obtain a test condition-component response characteristic fine-grained interaction coding vector, including: Based on external knowledge, the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedding coding vector are optimized for feature interaction based on an attention mechanism to obtain an external knowledge optimized test condition-component response characteristic fine-grained interaction feature matrix; Optimize the test condition-component response characteristic fine-grained interactive feature matrix based on the external knowledge, perform feature modulation optimization on the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedded coding vector respectively to obtain an optimized core component response characteristic significant aggregation coding vector and an optimized test condition low-dimensional embedded coding vector; The test condition-component response characteristic fine-grained interaction coding vector is obtained by calculating the point-by-point division between the optimized core component response characteristic significant aggregation coding vector and the optimized test condition low-dimensional embedding coding vector.

2. The meteorological instrument fault detection method according to claim 1, characterized in that: The set of the core component response characteristic data is dynamically aggregated based on embedded coding to obtain a core component response characteristic significant aggregation coding vector, including: Each core component response characteristic data in the set of core component response characteristic data is passed through an embedding layer to obtain a set of core component response characteristic semantic embedding coding vectors; Calculating a pseudo-anchored aggregation center representation vector of a core component response characteristic based on a characteristic static potential energy distribution of a set of semantic embedding encoding vectors of the core component response characteristic; Calculating the aggregated motion direction of each core component response characteristic semantic embedding coding vector in the set of the core component response characteristic semantic embedding coding vectors relative to the core component response characteristic pseudo-anchored aggregated center representation vector to obtain a set of core component response characteristic aggregated motion directions; Based on the set of aggregation movement directions of the core component response characteristics, the set of the core component response characteristic semantic embedding coding vectors is dynamically aggregated toward the core component response characteristic pseudo-anchor aggregation center representation vector to obtain the core component response characteristic significant aggregation coding vector.

3. The meteorological instrument fault detection method according to claim 2, characterized in that: Based on the characteristic static potential energy distribution of the set of the semantic embedding coding vectors of the core component response characteristics, a pseudo-anchored aggregation center representation vector of the core component response characteristics is calculated, including: Inputting each core component response characteristic semantic embedding coding vector in the set of core component response characteristic semantic embedding coding vectors into a characteristic static potential energy measurement network to obtain a set of core component response characteristic static potential energy measurement coefficients; Inputting the set of core component response characteristic static potential energy measurement coefficients into a characteristic energy level screening gating unit to obtain a set of core component response characteristic static potential energy weight factors; Based on the set of static potential weight factors of the core component response characteristics, a weighted sum of the set of semantic embedding coding vectors of the core component response characteristics is calculated to obtain a pseudo-anchored aggregation center representation vector of the core component response characteristics.

4. The meteorological instrument fault detection method according to claim 3, characterized in that: Inputting each core component response characteristic semantic embedding coding vector in the set of core component response characteristic semantic embedding coding vectors into a characteristic static potential energy measurement network to obtain a set of core component response characteristic static potential energy measurement coefficients, including: Performing Z-score normalization processing on the core component response characteristic semantic embedding coding vector to obtain a normalized core component response characteristic semantic embedding coding vector; The sum of the cube of each eigenvalue in the standardized core component response characteristic semantic embedding coding vector is calculated and divided by the characteristic scale value of the standardized core component response characteristic semantic embedding coding vector to obtain the core component response characteristic static potential energy measurement coefficient.

5. The meteorological instrument fault detection method according to claim 4, characterized in that: The core component response characteristic aggregation movement direction is the arccosine function value between the core component response characteristic semantic embedding coding vector and the core component response characteristic pseudo-anchor aggregation center representation vector.

6. The meteorological instrument fault detection method according to claim 5, characterized in that: Based on external knowledge, the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedding coding vector are optimized for feature interaction based on an attention mechanism to obtain an external knowledge optimized test condition-component response characteristic fine-grained interaction feature matrix, including: Inputting the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedding coding vector into a fine-grained feature interaction network to obtain a test condition-component response characteristic fine-grained interaction feature matrix; The test condition-component response characteristic fine-grained interaction feature matrix is ​​input into an attention unit based on external knowledge to obtain the external knowledge optimized test condition-component response characteristic fine-grained interaction feature matrix.

7. The meteorological instrument fault detection method according to claim 6, characterized in that: Optimizing the test condition-component response characteristic fine-grained interactive feature matrix based on the external knowledge, performing feature modulation optimization on the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedded coding vector respectively to obtain an optimized core component response characteristic significant aggregation coding vector and an optimized test condition low-dimensional embedded coding vector, including: Performing a linear transformation on the core component response characteristic significant aggregation coding vector to obtain a first query feature vector and a first value feature vector, and using the external knowledge optimization test condition-component response characteristic fine-grained interaction feature matrix as a key matrix, inputting the first query feature vector, the first value feature vector and the key matrix into a fine-grained modulation module based on a Transformer structure to obtain the optimized core component response characteristic significant aggregation coding vector; The test condition low-dimensional embedded coding vector is linearly transformed to obtain a second query feature vector and a second value feature vector, and the external knowledge-optimized test condition-component response characteristic fine-grained interaction feature matrix is ​​used as a key matrix. The second query feature vector, the second value feature vector and the key matrix are input into the Transformer-based fine-grained modulation module to obtain the optimized test condition low-dimensional embedded coding vector.

8. The meteorological instrument fault detection method according to claim 7, characterized in that: Determining whether a tested meteorological instrument fails based on the test condition-component response characteristic fine-grained interaction encoding vector includes: The test condition-component response characteristic fine-grained interactive coding vector is input into a classifier-based fault detection module to obtain a fault detection result, and the fault detection result is used to indicate whether the tested meteorological instrument has a fault.

9. A meteorological instrument fault detection system, characterized in that: include: A test condition setting module, used to set test conditions, wherein the test conditions include input voltage value, input current value and optical signal strength value; A response characteristic data acquisition module, used to acquire the response characteristic data of each core component of the tested meteorological instrument under the test conditions to obtain a set of core component response characteristic data; A test condition embedding coding module, used for embedding the test condition to obtain a low-dimensional embedding coding vector of the test condition; A feature dynamic aggregation processing module, used for performing feature dynamic aggregation based on embedded coding on the set of core component response characteristic data to obtain a core component response characteristic significant aggregation coding vector; A feature fine-grained interaction processing module, used for performing feature fine-grained interaction based on external knowledge enhancement on the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedded coding vector to obtain a test condition-component response characteristic fine-grained interaction coding vector; A meteorological instrument test result determination module, used to determine whether a tested meteorological instrument fails based on the test condition-component response characteristic fine-grained interactive coding vector; The feature fine-grained interaction processing module is used to: Based on external knowledge, the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedding coding vector are optimized for feature interaction based on an attention mechanism to obtain an external knowledge optimized test condition-component response characteristic fine-grained interaction feature matrix; Optimize the test condition-component response characteristic fine-grained interactive feature matrix based on the external knowledge, perform feature modulation optimization on the core component response characteristic significant aggregation coding vector and the test condition low-dimensional embedded coding vector respectively to obtain an optimized core component response characteristic significant aggregation coding vector and an optimized test condition low-dimensional embedded coding vector; The test condition-component response characteristic fine-grained interaction coding vector is obtained by calculating the point-by-point division between the optimized core component response characteristic significant aggregation coding vector and the optimized test condition low-dimensional embedding coding vector.

Images