Terminal temperature monitoring and fault automatic diagnosis method of intelligent measuring switch

Abstract

The application provides a terminal temperature monitoring and automatic fault diagnosis method of a smart metering switch, relates to the technical field of fault diagnosis, and comprises the following steps: performing dynamic tracking on the terminal temperature to obtain long-term and short-term temperature time-varying characteristics; if preset temperature abnormality judgment conditions are met, starting and ending time stamps are called; local multi-source monitoring data is called to obtain multi-source time sequence monitoring data; the multi-source monitoring data is divided into P monitoring data segments; segmented fault diagnosis is performed based on dynamic fluctuation characteristics, P initial fault prediction results are output; state transition probability correction is performed to obtain P-1 updated fault prediction results; time characteristic fusion decision is performed to output a fault to be intervened and trigger fault grading early warning processing. The application solves the technical problem that the prior art usually only judges the fault state of the smart metering switch through static fault mode recognition, cannot respond or adjust intervention measures in a timely manner in the fault development process, and thus increases fault diagnosis delay and false alarm risk.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, specifically to a method for terminal temperature monitoring and automatic fault diagnosis of intelligent measuring switches. Background Technology

[0002] Intelligent metering switches are circuit breakers that support local and remote communication settings for data and parameters. They possess multiple IoT sensing functions, including HPLC communication, high-precision AC input, power supply topology identification, metering box management, electricity theft warning, power outage reporting, line loss analysis and judgment, terminal temperature detection, and automatic fault diagnosis. They also offer various protection functions such as over / under voltage protection, phase loss protection, long-delay overload protection, short-delay short-circuit protection, and instantaneous short-circuit protection. Intelligent metering switches play a crucial role in detecting and protecting lines and power equipment, reducing grid maintenance costs, and providing reliable data for energy efficiency management systems. However, the stable operation of intelligent metering switches directly depends on the normal operating status of their components, especially terminal temperature. Excessive temperature often leads to a decrease in the insulation performance of equipment, resulting in electrical faults or equipment damage. Therefore, terminal temperature monitoring and automatic fault diagnosis are key technologies in intelligent metering switches. However, traditional technologies typically rely solely on static fault mode identification to determine the fault status of smart measurement switches. Furthermore, when multiple fault modes occur simultaneously, it is often difficult to accurately predict the fault status, which in turn leads to an inability to respond or adjust intervention measures in a timely manner during the fault development process, thereby increasing diagnostic delays and the risk of false alarms. Summary of the Invention

[0003] This application provides a terminal temperature monitoring and automatic fault diagnosis method for intelligent measuring switches, aiming to solve the technical problem that existing technologies typically rely solely on static fault mode identification to determine the fault status of intelligent measuring switches, which cannot respond or adjust intervention measures in a timely manner during the fault development process, thereby increasing the risk of fault diagnosis delay and false alarm.

[0004] This application discloses a method for terminal temperature monitoring and automatic fault diagnosis of an intelligent measuring switch. The method includes: dynamically tracking terminal temperature based on a sliding monitoring window to obtain long-term and short-term temperature time-varying characteristics; if the long-term and short-term temperature time-varying characteristics meet preset temperature anomaly judgment conditions, retrieving the start and end timestamps of the long-term and short-term temperature time-varying characteristics; calling local multi-source monitoring data of the intelligent measuring switch according to the start and end timestamps to obtain multi-source time-series monitoring data; dividing the multi-source time-series monitoring data into P multi-source monitoring data segments using the 1 / P sliding monitoring window; performing segmented fault diagnosis based on the dynamic fluctuation characteristics of the P multi-source monitoring data segments, and outputting P initial fault prediction results; correcting the state transition probability of the P initial fault prediction results to obtain P-1 updated fault prediction results; aligning the P initial fault prediction results and P-1 updated fault prediction results to perform time feature fusion decision-making, outputting the fault to be intervened, and triggering fault classification early warning processing.

[0005] One or more technical solutions provided in this application have at least the following beneficial effects:

[0006] By dynamically tracking terminal temperatures through a sliding monitoring window, the short-term and long-term temperature variation characteristics of terminals can be tracked and analyzed in real time. This method can promptly identify abnormal temperature fluctuations in equipment and monitor potential fault risks, especially sudden temperature changes or continuous temperature increases. Extracting short-term and long-term temperature time-varying characteristics helps to more comprehensively capture temperature fluctuation patterns, detecting not only rapid temperature changes in a short period but also identifying continuous temperature increases. By retrieving start and end timestamps, the occurrence and end times of temperature anomalies can be accurately located. This precise anomaly time marking provides a critical time window for subsequent data analysis and fault diagnosis, helping to further analyze the root cause of the fault. Based on the start and end timestamps, relevant time-series data is extracted from local multi-source monitoring data, ensuring that the data used is accurate and relevant to the abnormal period. This operation ensures data relevance, improves the reliability of fault diagnosis, and avoids interference from irrelevant data. By dividing multi-source time-series monitoring data into P data segments, the data can be broken down into smaller time intervals. The data is segmented, allowing for detailed analysis of each segment individually. Segmented processing improves the accuracy and efficiency of data processing. Segmented fault diagnosis based on the dynamic fluctuation characteristics of each data segment effectively captures the fluctuation patterns of the equipment under different operating conditions. This method can identify abnormal fluctuations or potential faults in the equipment and output corresponding initial fault prediction results. State transition probability correction of the initial fault prediction results considers the actual probability of the equipment fault transitioning from one state to another, thereby optimizing fault prediction. State transition probability correction more accurately reflects the actual operating state of the equipment and eliminates erroneous judgments caused by prediction bias or data errors. By integrating the initial and updated fault prediction results through time feature fusion, a final fault judgment is obtained. This decision-making mechanism can balance the continuity of time series data with the accuracy of fault prediction, reducing the misdiagnosis rate. Fault classification early warning processing is triggered, and corresponding intervention measures are formulated according to the severity of the fault, effectively reducing the risks caused by equipment failure.

[0007] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0008] Figure 1 A schematic flowchart of a method for terminal temperature monitoring and automatic fault diagnosis of an intelligent measuring switch is provided for embodiments of this application.

[0009] Figure 2 This application provides a schematic diagram of the segmented fault diagnosis process in the terminal temperature monitoring and automatic fault diagnosis method of an intelligent measuring switch. Detailed Implementation

[0010] This application provides a terminal temperature monitoring and automatic fault diagnosis method for intelligent measuring switches, which solves the technical problem that the existing technology usually judges the fault status of intelligent measuring switches only through static fault mode identification, and cannot respond or adjust intervention measures in a timely manner during the fault development process, thereby increasing the delay in fault diagnosis and the risk of false alarms.

[0011] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0012] like Figure 1 As shown in the embodiment of this application, a method for terminal temperature monitoring and automatic fault diagnosis of an intelligent measuring switch is provided, the method comprising:

[0013] By dynamically tracking the terminal temperature using a sliding monitoring window, the time-varying characteristics of long-term and short-term temperature are obtained.

[0014] The sliding monitoring window is a dynamic window used to monitor terminal temperature changes. By acquiring terminal temperature data at set time intervals, the sliding monitoring window can continuously collect temperature data and capture the dynamic changes in terminal temperature, thereby obtaining long-term and short-term temperature time-varying characteristics. Long-term and short-term temperature time-varying characteristics refer to the characteristics of temperature change at different time scales. Among them, short-term characteristics capture temperature fluctuations in a short period of time (e.g., minutes or hours), such as a sudden rise or a sharp drop in temperature; long-term characteristics capture temperature change trends in a short period of time (e.g., days or weeks), which are used to identify temperature stability or a gradually increasing trend.

[0015] If the long-term and short-term temperature time-varying characteristics meet the preset temperature anomaly judgment conditions, then the start and end timestamps of the long-term and short-term temperature time-varying characteristics are retrieved.

[0016] The preset temperature anomaly judgment conditions include transient thermal stress critical rate, insulation degradation temperature rise gradient threshold, and spatial thermal imbalance deviation limit. When the long-term and short-term temperature time-varying characteristics of the terminal temperature reach or exceed these conditions, it is judged that an anomaly has occurred. At this time, the start and end time points of data acquisition are extracted according to the sliding monitoring window that collects long-term and short-term temperature time-varying characteristics, i.e., start and end timestamps.

[0017] Based on the start and end timestamps, the local multi-source monitoring data of the intelligent measurement switch is retrieved to obtain multi-source time-series monitoring data.

[0018] Based on the start and end timestamps, all relevant monitoring data within that time period is retrieved. Local multi-source monitoring data includes monitoring data from multiple data sources besides terminal temperature, such as current, voltage, and humidity, which can all affect the working status of the intelligent measuring switch. Multi-source time-series monitoring data consists of time-series data from multiple data sources, meaning that this data is arranged in chronological order, showing the trends of each data source over time.

[0019] The multi-source time-series monitoring data is divided into P multi-source monitoring data segments using the 1 / P sliding monitoring window.

[0020] The sliding monitoring window is used for data processing. However, in this step, the multi-source time-series monitoring data is divided into segments equal to 1 / P of the sliding monitoring window, where P is a positive integer. The segmentation coefficient P is determined based on the total length of the data and the sliding monitoring window. Assuming the sliding monitoring window length is T, the time range of each data segment will be half of the original, i.e., T / P. This is to further subdivide the data for more precise subsequent analysis. The multi-source time-series monitoring data is divided into multiple smaller data segments according to the time period, resulting in P multi-source monitoring data segments. Each multi-source monitoring data segment contains monitoring information from different data sources.

[0021] Based on the dynamic fluctuation characteristics of the P multi-source monitoring data segments, segmented fault diagnosis is performed, and P initial fault prediction results are output.

[0022] Each multi-source monitoring data segment has corresponding dynamic fluctuation characteristics. These characteristics refer to the patterns of data change over time, such as fluctuation amplitude and frequency. These dynamic fluctuation characteristics can reveal the health status or fault modes of equipment operation. For example, using methods such as Fourier transform, time series analysis, and volatility analysis, the dynamic fluctuation characteristics of each multi-source monitoring data segment can be extracted to identify whether there are abnormal conditions in the equipment, such as vibration or temperature fluctuations. Based on the dynamic fluctuation characteristics of each multi-source monitoring data segment, fault diagnosis is performed to determine the existence and type of fault. Each multi-source monitoring data segment outputs an initial fault prediction result, indicating whether there are potential faults or abnormalities in that segment.

[0023] The state transition probability is corrected for the P initial fault prediction results to obtain P-1 updated fault prediction results.

[0024] In fault prediction, changes in system state need to be considered. That is, a fault mode may transition over time or under different conditions. State transition probability correction represents the probability of transitioning from one fault state to another. A state transition probability matrix is ​​constructed using historical fault data and pattern recognition, representing the probability of the device transitioning from one fault mode to another. The initial fault prediction results are corrected using the state transition probability matrix to obtain updated fault prediction results. Since each data segment corresponds to a time window, after state transition probability correction, each of the P initial prediction results will be adjusted based on the previous result, ultimately outputting P-1 updated fault prediction results. For example, if the initial fault prediction indicates that the device may experience a certain fault, the state transition correction will determine whether the fault will continue to develop or transform into another fault mode based on historical data and transition probabilities.

[0025] Align the P initial fault prediction results and P-1 updated fault prediction results with time feature fusion decision, output the fault to be intervened, and trigger fault classification early warning processing.

[0026] By combining P initial fault prediction results and P-1 updated fault prediction results, a fusion analysis is performed. Since the fault prediction results are based on the analysis of different data segments, a time feature fusion algorithm is used to integrate data from multiple time periods to obtain a more comprehensive and accurate fault warning. Through time feature fusion, a final decision result is output, indicating which faults require immediate intervention. By combining fault prediction information from different time periods, it is possible to determine which fault modes have a higher risk, thereby triggering warnings and taking corresponding intervention measures. Based on the output fault prediction results, a fault classification warning is triggered. Faults are classified into multiple levels according to their severity and impact. The warning system takes different countermeasures based on this classification result. For example, some severe faults require immediate shutdown, while other minor faults only require further monitoring.

[0027] Furthermore, such as Figure 2 As shown, segmented fault diagnosis is performed based on the dynamic fluctuation characteristics of the P multi-source monitoring data segments, and P initial fault prediction results are output. The method includes:

[0028] Dynamic fluctuation features are extracted from the first multi-source monitoring data segment to obtain multi-source dynamic fluctuation indicators; the multi-source dynamic fluctuation indicators are combined to construct a first fault feature vector; multiple sets of sample-level fault feature vectors of various sample fault modes are locally invoked; the various sample fault modes and the multiple sets of sample-level fault feature vectors are associated and stored to construct a fault mode library; the first fault feature vector is used to perform fault mode recognition in the fault mode library to obtain a first set of fault level confidence scores; the first set of fault level confidence scores is subjected to probability normalization processing to output the first initial fault prediction result.

[0029] The first multi-source monitoring data segment is any one of the P multi-source monitoring data segments and is used as the current analysis object. The time series characteristics of the first multi-source monitoring data segment are analyzed to extract the dynamic fluctuation characteristics of the data. These dynamic fluctuation characteristics reflect changes in the equipment's state during operation and the presence of potential fault signs. These characteristics include: amplitude fluctuations (the amplitude of data fluctuations can reflect abnormal conditions of the equipment; for example, sharp fluctuations in temperature, pressure, or current may indicate a fault or a precursor to a fault); frequency fluctuations (frequent fluctuations in the data indicate vibration, resonance, or other periodic faults in the equipment); and the rate of change of fluctuation amplitude (the speed at which the fluctuation amplitude changes can also reveal whether the equipment is rapidly entering a fault state). The combination of data from different sensors can provide a comprehensive picture of the equipment's state. Each monitoring source has different fluctuation characteristics; therefore, extracting these dynamic fluctuation characteristics from each monitoring source can provide more information for fault diagnosis. The resulting multi-source dynamic fluctuation indicators reflect the fluctuation situation of various aspects of the equipment and serve as the foundational data for subsequent fault diagnosis.

[0030] The extracted multi-source dynamic fluctuation indicators are combined into a first fault feature vector. The first fault feature vector is formed by concatenating the fluctuation indicators of each monitoring source. The purpose is to summarize the features of each monitoring source into a multi-dimensional vector. Each dimension represents the dynamic fluctuation feature of a monitoring source. In this way, a preliminary feature vector representing the equipment fault state is constructed, which represents the operating status of the equipment within the time period, thereby providing a basis for fault diagnosis.

[0031] The system locally invokes multiple sample fault modes. Each sample fault mode includes monitoring data characteristics of the equipment under that fault state, such as the change patterns of indicators like temperature, vibration, and current. These sample fault modes provide a reference for fault diagnosis of the equipment status. Each sample fault mode has a corresponding set of sample-level fault feature vectors, which are generated from historical or simulated data and represent the equipment's behavior patterns under different fault levels.

[0032] By associating and storing multiple sample fault modes and their corresponding multiple sets of sample-level fault feature vectors, a fault mode library is constructed. The fault mode library is a database containing various fault mode samples and their features, which can be used to compare the similarity between the current equipment status and historical fault modes.

[0033] Using the first fault feature vector, a matching process is performed in the fault mode library to calculate the similarity between feature vectors. This identifies which historical fault modes the current data segment is most similar to. Similarity calculation methods include Euclidean distance and cosine similarity. By calculating the similarity between the current feature vector and the feature vector of each fault mode in the fault mode library, it can be determined which fault mode is closest to the current equipment state. The result of fault mode recognition not only identifies the best-matching fault mode but also provides a confidence score for each fault mode. The confidence score reflects the probability of each fault mode at different levels and is usually a probability distribution representing the likelihood of the fault mode occurring.

[0034] The first set of fault classification confidence scores obtained from fault mode identification is normalized. Normalization transforms the confidence scores of multiple fault modes into a standardized probability distribution, ensuring that the confidence scores of all fault modes are between 0 and 1, and that the sum of all confidence scores is 1. After normalization, a standardized confidence distribution is obtained, and the fault mode with the highest confidence score and its corresponding severity level are extracted as the final prediction result, i.e., the first initial fault prediction result.

[0035] Furthermore, the confidence scores of the first group of fault classifications are subjected to probability normalization to output the first initial fault prediction result. The method includes:

[0036] The first set of fault classification confidence scores is mutually exclusive, and multiple fault mode pairs with logical conflicts are located. Based on the priority of the physical contradiction evidence strength of the multi-source dynamic fluctuation index, the confidence weights of the multiple fault mode pairs are reset by traversing the first set of fault classification confidence scores, and the first set of reset classification confidence scores is output. The first set of reset classification confidence scores is subjected to probability normalization calculation to generate a normalized confidence probability distribution. The first initial fault prediction result is extracted from the normalized confidence probability distribution.

[0037] Check whether there is a logical conflict in the first set of fault classification confidence scores obtained from fault mode recognition. A logical conflict means that two or more fault modes cannot occur at the same time, but fault mode recognition gives contradictory fault mode predictions. If a conflict is found in the first set of fault classification confidence scores, for example, the probability values ​​of two fault modes are both high, but they are mutually exclusive, then locate these conflict pairs, i.e., fault mode pairs. Each fault mode pair represents the situation where two fault modes contradict each other at the current time.

[0038] Based on the physical evidence from the multi-source dynamic fluctuation index, the strength of physical contradiction evidence for different failure modes in each data source is calculated. This refers to whether each data source strongly supports a certain failure mode and reflects the strength of this supporting evidence. For multiple failure mode pairs, the confidence of conflicting failure mode pairs is adjusted according to the strength of physical contradiction evidence. This is done by resetting the confidence of each failure mode pair by assigning different weights. If a failure mode has stronger physical evidence support, its confidence will be enhanced, while the confidence of another failure mode will be reduced. After the above confidence reset, the first set of reset graded confidence scores is output, reflecting the correction result of the failure modes based on the strength of physical evidence.

[0039] The confidence scores of the first set of reset levels are subjected to probability normalization. Normalization transforms the confidence scores of multiple failure modes into a standardized probability distribution, ensuring that the sum of all confidence scores is 1, thus making the confidence score of each failure mode a probability. The result of the normalization process is a normalized confidence probability distribution, representing the confidence score for each failure mode based on current monitoring data and historical experience. The normalized probability distribution provides a standardized basis for the final assessment of failure modes.

[0040] The fault mode with the highest confidence level, i.e., the fault mode with the highest probability value, is extracted from the normalized confidence probability distribution. This fault mode is the most likely type of fault under the current equipment state. Based on the set confidence threshold, it is checked whether the confidence level of the extracted fault mode reaches a specific minimum level. Only when the confidence level of the fault mode exceeds this confidence threshold is it considered a valid prediction result. When extracting fault modes, it is ensured that fault modes without logical conflicts are selected. Finally, the first initial fault prediction result is obtained, which is the highest confidence fault mode and its severity level that meet the preset confidence threshold and have no logical conflicts.

[0041] Furthermore, based on the priority of the physical contradiction evidence strength of the multi-source dynamic fluctuation index, the confidence weights of the multiple fault mode pairs are reset by traversing the first group of fault classification confidence scores, and the first group of reset classification confidence scores is output. The method includes:

[0042] A pre-constructed physical contradiction evidence priority rule base is established; multiple fault classification confidence pairs of the multiple fault mode pairs are retrieved from the first group of fault classification confidence scores; the multi-source dynamic fluctuation index and the multiple fault classification confidence pairs are used as joint retrieval conditions, and the physical contradiction evidence priority rule base is traversed to retrieve multiple matching confidence reset operators; after performing confidence reset operations on the multiple fault classification confidence pairs using the matching confidence reset operators, the first group of fault classification confidence scores is replaced and updated according to the reset results, and the first group of reset classification confidence scores is output.

[0043] Physical conflict evidence prioritization refers to determining the credibility of each failure mode based on the strength of evidence from different physical data sources when multiple failure mode predictions conflict or contradict each other. Some physical data sources provide stronger support for a certain failure mode, while other data sources may provide weaker support. In order to make effective decisions when failure modes conflict, a physical conflict evidence prioritization rule base is pre-built. This rule base contains a series of rules and priorities, which can determine the priority of each failure mode according to the strength of evidence from different physical data sources when conflicts occur.

[0044] From the first set of fault classification confidence scores, fault mode pairs involving multiple fault modes are extracted. Each fault mode pair represents the situation where two possible fault modes conflict within the same time window. These fault modes are logically mutually exclusive. For example, the probability of two fault modes coexisting is zero. It is necessary to identify these fault mode pairs and retrieve the corresponding confidence scores from these fault mode pairs to represent the system's confidence in the fault mode, thus forming fault classification confidence score pairs.

[0045] This method combines multi-source dynamic fluctuation indicators with multiple fault level confidence pairs as joint retrieval conditions to query the physical conflict evidence priority rule base. Joint retrieval means considering not only the confidence of the fault mode but also the physical evidence from multiple source data to find the best match among multiple conflicting fault modes. The rule base is traversed to find priority rules that meet the conditions, and the credibility of each fault mode is determined based on these priority rules. Based on these priority rules, the matching confidence reset operator is invoked to adjust the confidence of the fault mode so that it can be corrected based on physical evidence in case of conflict.

[0046] The matching confidence reset operator adjusts multiple fault grade confidence pairs. The reset operation adjusts the confidence of conflicting fault modes based on the strength of physical evidence. For example, when the physical evidence strongly supports a fault mode, its confidence is increased, while the confidence of conflicting fault modes is decreased. This confidence reset ensures that the confidence of each fault mode better matches the support provided by the physical data source. After the confidence reset operation, the first set of fault grade confidences is updated to reflect the reassessment of the fault modes. This updated first set of reset confidence levels serves as the basis for fault prediction and is used for subsequent fault mode identification and severity assessment.

[0047] Furthermore, the methods also include:

[0048] Based on the application scenario and model ID of the intelligent measurement switch, network data is retrieved to obtain multiple sample time-series fault transfer sequences; based on the multiple sample time-series fault transfer sequences, a frequent pattern mining algorithm is applied to obtain a valid fault transfer path; the transfer probability density of the multiple sample time-series fault transfer sequences is estimated to obtain multiple sets of associated fault transfer probabilities for various sample fault types; after constructing a fault state transfer probability matrix based on the various sample fault types, the fault state transfer probability matrix is ​​filled with the multiple sets of associated fault transfer probabilities.

[0049] Data is retrieved from the network based on the application scenario and model ID of the intelligent measurement switch. Different application scenarios impose different workloads, environmental factors, and failure modes on the equipment. For example, equipment used in a high-temperature environment exhibits different failure transfer modes than equipment used in a normal-temperature environment. Different models of intelligent measurement switches have different hardware and software configurations, resulting in differences in their failure modes and transfer paths. Appropriate datasets are selected based on the model ID. Multiple sample time-series failure transfer sequences are obtained through network access. These sequences contain the failure state transfer information of the equipment during operation, i.e., the time sequence and transfer conditions of the equipment transitioning from one failure state to another. Each sample sequence represents the failure transfer trajectory of the equipment over a period of time.

[0050] Frequent pattern mining is a data mining technique designed to discover frequently occurring patterns in large datasets. In this step, frequent pattern mining analyzes multiple sample time-series fault transition sequences to extract frequently occurring fault transition patterns. Each time-series fault transition sequence contains transitions between fault states. By mining these sequences, legitimate fault transition paths that frequently occur in multiple samples can be identified. These paths represent the common order in which a fault transitions from one state to another. Through frequent pattern mining, legitimate fault transition paths are identified. Legitimate fault transition paths are those state transitions that occur frequently in multiple samples, reflecting the normal or expected fault evolution process of the equipment.

[0051] Transition probability density estimation describes the distribution of transition probability from one fault state to another. By analyzing multiple sample time-series fault transition sequences, the transition probability between each pair of fault states is calculated. These probabilities represent the likelihood of transitioning from one fault state to another. Each equipment fault type may have different transition patterns and probabilities. By estimating the transition probabilities of multiple sample time-series fault transition sequences, the associated fault transition probabilities for different sample fault types are obtained. The associated fault transition probabilities reflect the transition relationships between different fault types and provide data support for the subsequent construction of the fault state transition matrix.

[0052] The fault state transition probability matrix is ​​a matrix representing the transition probabilities between different fault states. Each matrix element represents the transition probability from one fault state to another. Through the fault state transition probability matrix, we can understand the relationships between fault states and the likelihood of their transition. Based on the various sample fault types calculated above, a complete fault state transition probability matrix is ​​constructed. This matrix is ​​then filled with multiple sets of associated fault transition probabilities, and the filled values ​​reflect the actual transition probabilities between different fault states. The fault state transition probability matrix provides crucial information for equipment fault prediction, condition monitoring, and early warning systems.

[0053] Furthermore, the state transition probability is corrected on the P initial fault prediction results to obtain P-1 updated fault prediction results. The method includes:

[0054] Extract a first predicted fault mode from the first initial fault prediction result; extract a second predicted fault mode from the second initial fault prediction result; using the first predicted fault mode as the transition starting state, load the second predicted fault mode into the fault state transition probability matrix, and query and output the second transition probability weight; use the second transition probability weight to correct the initial confidence of the second initial fault prediction result to obtain the second updated fault prediction result; similarly perform state transition probability correction on the P initial fault prediction results to obtain the P-1 updated fault prediction results.

[0055] Extract the predicted fault modes from the two initial fault prediction results. These two initial fault prediction results correspond to the prediction results of two different time periods. The first predicted fault mode and the second predicted fault mode are the fault types that the equipment is most likely to encounter in these two different time periods.

[0056] The first predicted fault mode is used as the starting point for the fault state transition. That is, the equipment is initially in a specific fault mode, such as excessive temperature or abnormal vibration. The second predicted fault mode is used as the target state. The transition probability from the first predicted fault mode to the second predicted fault mode is found in the fault state transition probability matrix. The weight of the second transition probability is output, which represents the possibility of transitioning from the first predicted fault mode to the second predicted fault mode.

[0057] The initial confidence level of the second initial fault prediction result is adjusted based on the weight of the second transition probability. Specifically, the transition probability from the first fault mode to the second fault mode is applied to the confidence level of the second initial fault prediction result for adjustment. If the weight of the second transition probability is higher, the confidence level of the second predicted fault mode will increase accordingly; if the weight of the second transition probability is lower, the confidence level of the second predicted fault mode will decrease. This adjustment is based on the probability of the device transitioning from its current state to a future fault state. After the transition probability adjustment, the second updated fault prediction result is obtained, which indicates how the device will transition to the next fault mode based on the current device state and fault mode.

[0058] Each initial fault prediction result is sequentially corrected for state transition probabilities. Specifically, based on the transition probability of the previous fault mode, the confidence of each subsequent fault mode is sequentially corrected until the last prediction result. Each transition correction adjusts the confidence of the next fault mode based on the transition probability of the previous fault mode. By correcting the state transition probabilities of P initial fault prediction results, P-1 updated fault prediction results are finally obtained. These updated results reflect the transition of the equipment from the initial state to each subsequent state and provide a basis for fault intervention and early warning processing.

[0059] Furthermore, if the fault transfer chain from the first predicted fault mode to the second predicted fault mode does not satisfy the legal fault transfer path, then the second updated fault prediction result is rolled back to the second initial fault prediction result.

[0060] The fault transfer chain from the first predicted fault mode to the second predicted fault mode is checked to ensure it follows a valid fault transfer path. The fault transfer chain represents the sequence and path by which the device transitions from one fault mode to another. During normal operation and fault mode transitions, some transitions are valid, while others are illogical or inconsistent with the device's physical characteristics. If the fault transfer chain from the first predicted fault mode to the second predicted fault mode does not conform to the predetermined valid path, the second updated fault prediction result is rolled back, restoring it to the second initial fault prediction result. In other words, fault prediction results obtained through invalid paths are discarded, and the system returns to the initial prediction state. This process helps maintain system stability and avoids false alarms or incorrect decisions caused by illogical fault mode transitions.

[0061] Furthermore, if the long-term and short-term temperature time-varying characteristics meet the preset temperature anomaly determination conditions, then the start and end timestamps of the long-term and short-term temperature time-varying characteristics are retrieved, including:

[0062] The short-term and long-term temperature time-varying features are decomposed to obtain the short-term temperature change rate, the long-term temperature rise trend slope, and the standard deviation of adjacent terminal temperatures. The temperature anomaly judgment conditions are used to iterate through the short-term temperature change rate, the long-term temperature rise trend slope, and the standard deviation of adjacent terminal temperatures. When any feature continuously meets the anomaly condition, the first abnormal data sampling time is located back. The last abnormal data sampling time is retrieved from the short-term and long-term temperature time-varying features. The first abnormal data sampling time and the last abnormal data sampling time constitute the start and end timestamps.

[0063] Equipment temperature data exhibits short-term and long-term trends. Short- and long-term temperature time-varying characteristics refer to the temperature changes of the equipment at different time scales. Analyzing these characteristics allows for a better understanding of the equipment's operating status and helps determine if temperature anomalies exist. Specifically, the short-term temperature change rate reflects the speed of temperature change within a short period. This indicator can capture rapid temperature fluctuations and is often used to detect the risk of instantaneous overheating or sudden temperature changes. The short-term temperature change rate is obtained by calculating the temperature difference between adjacent time points. The long-term temperature rise trend slope represents the trend of temperature change over a longer period. It can detect whether the equipment is experiencing continuous temperature increases, a key indicator for determining potential equipment malfunctions. The long-term temperature rise trend slope can be calculated using linear regression analysis of the slope of temperature change over time. The standard deviation of adjacent terminal temperatures reflects the degree of temperature fluctuation between different terminals of the equipment. A large standard deviation indicates uneven temperature distribution, potentially indicating thermal imbalance or localized overheating. The standard deviation of adjacent terminal temperatures is obtained by calculating the standard deviation of multiple terminal temperatures.

[0064] The system iterates through the short-term temperature change rate, the long-term temperature rise trend slope, and the standard deviation of adjacent terminal temperatures in sequence, and applies the above-mentioned temperature anomaly judgment conditions. If any feature continuously exceeds its threshold and meets the anomaly conditions, the equipment is considered to have a temperature anomaly. When a certain temperature feature of the equipment exceeds the anomaly judgment conditions, the temperature data is backtracked to locate the first time the abnormal data was sampled. This refers to the time point when the equipment first experienced a temperature anomaly.

[0065] Based on the time-varying characteristics of long and short-term temperature, the end point of the anomaly is determined, i.e. the last time the abnormal data was sampled. This means that the temperature data of the equipment continues to change under abnormal conditions until the last sampling time. By determining the first time the abnormal data was sampled and the last time the abnormal data was sampled, the time period of the temperature anomaly is determined, i.e. the start and end timestamps. This time period represents the complete abnormal process of the equipment from the first occurrence of temperature anomaly to the last sampling.

[0066] Furthermore, the temperature anomaly determination criteria include the transient thermal stress critical rate, the insulation degradation temperature rise gradient threshold, and the spatial thermal imbalance deviation limit.

[0067] The transient thermal stress critical rate refers to the rate of temperature change within a short period of time. Exceeding a certain critical value may indicate an abnormal increase in thermal stress inside the equipment, which may lead to failure or damage. Rapid temperature changes may cause thermal expansion or contraction of materials, increasing the risk of equipment failure. The insulation degradation temperature rise gradient threshold assesses the degradation of the equipment's insulation materials under prolonged high-temperature operation. Insulation degradation leads to a decrease in the equipment's insulation performance, which may in turn lead to electrical failures or safety hazards. The temperature rise gradient represents the rate of temperature change. If the temperature rise is too fast, it may exacerbate the degradation of the insulation materials. The spatial thermal imbalance deviation limit assesses the temperature non-uniformity of different parts inside the equipment. Non-uniform temperature distribution may lead to local overheating, thermal stress concentration, or material deformation, increasing the risk of failure. The spatial thermal imbalance deviation represents the temperature difference between different parts of the equipment.

[0068] Furthermore, the method also includes:

[0069] The local maintenance record of the intelligent measurement switch is retrieved to obtain N fault maintenance data, where each fault maintenance data includes a fault maintenance difficulty coefficient and a fault maintenance effectiveness coefficient, and N is a positive integer; based on the N fault maintenance data, a fault maintenance quality assessment is performed to obtain a fault maintenance quality index; when the fault maintenance quality index is lower than a preset quality standard, the intelligent measurement switch is replaced.

[0070] Relevant fault maintenance data is retrieved from the local maintenance records of the intelligent measurement switch. Each fault maintenance data record records the specific handling of a fault, including the fault maintenance difficulty coefficient and the fault maintenance effectiveness coefficient. Here, N is a positive integer. The fault maintenance difficulty coefficient represents the technical complexity and workload required to repair the fault. The fault maintenance difficulty coefficient considers various factors, such as the severity of the fault, the required tools or spare parts, and the skill level of the technicians. The fault maintenance effectiveness coefficient represents the effect and degree of recovery after repair, which can be measured by factors such as the recovery status of the equipment and the degree of improvement in equipment performance.

[0071] A quality assessment is conducted based on the extracted N fault maintenance data points. For example, the assessment evaluates whether the maintenance difficulty matches the actual results; if a high-difficulty fault repair fails to achieve effective recovery, it indicates poor maintenance quality. The assessment also evaluates whether spare parts, personnel, and tools were used appropriately during the maintenance process, and whether maintenance was performed promptly and effectively reduced the impact time of the fault. Through these assessments, a fault maintenance quality index is assigned to the fault maintenance process of the intelligent measurement switch. This index is a comprehensive score representing the overall quality of equipment fault maintenance.

[0072] The fault maintenance quality index is compared with a preset quality standard, which is a benchmark value set based on historical maintenance data, industry standards, or equipment operating requirements, representing the minimum quality requirements for equipment maintenance. If the fault maintenance quality index is lower than the preset quality standard, the replacement operation of the intelligent measurement switch is initiated. This means that the fault of the intelligent measurement switch cannot be effectively repaired by existing maintenance methods, or the repair effect cannot meet the requirements for normal equipment operation. Therefore, the equipment needs to be replaced to ensure that the system can continue to operate stably.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A terminal temperature monitoring and fault automatic diagnosis method of a smart metering switch, characterized in that, The method comprises: Based on the sliding monitoring window, terminal temperature dynamic tracking is performed to obtain long-term and short-term temperature time-varying characteristics; If the long-term and short-term temperature time-varying characteristics meet the preset temperature abnormality judgment condition, the start and end time stamps of the long-term and short-term temperature time-varying characteristics are called; According to the start and end time stamps, the local multi-source monitoring data of the intelligent measurement switch is called to obtain multi-source time sequence monitoring data; The multi-source time sequence monitoring data is divided into P multi-source monitoring data segments by using 1 / P sliding monitoring window; Based on the dynamic fluctuation characteristics of the P multi-source monitoring data segments, segmented fault diagnosis is performed to output P initial fault prediction results; The state transition probability of the P initial fault prediction results is corrected to obtain P-1 updated fault prediction results; The time feature fusion decision of the P initial fault prediction results and P-1 updated fault prediction results is performed to output the fault to be intervened, and the fault grading early warning processing is triggered; The method further comprises: According to the application scene and model ID of the intelligent measurement switch, network data calling is performed to obtain a plurality of sample time sequence fault transition sequences; Based on the plurality of sample time sequence fault transition sequences, a frequent pattern mining algorithm is processed to obtain a fault legal transition path; The transition probability density of the plurality of sample time sequence fault transition sequences is estimated to obtain a plurality of groups of associated fault transition probabilities of a plurality of sample fault types; After the fault state transition probability matrix is constructed based on the plurality of sample fault types, the plurality of groups of associated fault transition probabilities are used to fill the fault state transition probability matrix.

2. The terminal temperature monitoring and fault automatic diagnosis method of the smart measurement switch according to claim 1, wherein, Based on the dynamic fluctuation characteristics of the P multi-source monitoring data segments, segmented fault diagnosis is performed to output P initial fault prediction results, which comprises: Dynamic fluctuation characteristics of the first multi-source monitoring data segment are extracted to obtain multi-source dynamic fluctuation indexes; The first fault feature vector is constructed by combining the multi-source dynamic fluctuation indexes; A plurality of groups of sample level fault feature vectors of a plurality of sample fault modes are called locally; The plurality of groups of sample level fault feature vectors of the plurality of sample fault modes are stored in association to construct a fault mode library; The first fault feature vector is used to perform fault mode recognition in the fault mode library to obtain a first group of fault grading confidence degrees; The first group of fault grading confidence degrees is subjected to probability normalization processing to output a first initial fault prediction result.

3. The terminal temperature monitoring and fault automatic diagnosis method of the smart measurement switch according to claim 2, wherein, The first group of fault grading confidence degrees is subjected to probability normalization processing to output a first initial fault prediction result, which comprises: The first group of fault grading confidence degrees is subjected to mutual exclusion verification to locate a plurality of fault mode pairs with logical conflicts; According to the priority of the physical contradiction evidence strength of the multi-source dynamic fluctuation indexes, the first group of fault grading confidence degrees is traversed to perform confidence weight resetting of the plurality of fault mode pairs to output a first group of reset grading confidence degrees; The first group of reset grading confidence degrees is subjected to probability normalization calculation to generate a normalized confidence probability distribution; The first initial fault prediction result is extracted from the normalized confidence probability distribution.

4. The terminal temperature monitoring and fault automatic diagnosis method of the smart measurement switch according to claim 3, wherein, According to the physical contradiction evidence strength priority of the multi-source dynamic fluctuation indicator, the confidence weight resetting of the plurality of fault mode pairs is performed on the first set of fault hierarchical confidence, and a first set of reset hierarchical confidence is output, the method comprising: pre-building a physical contradiction evidence priority rule library; calling a plurality of fault hierarchical confidence pairs of the plurality of fault mode pairs from the first set of fault hierarchical confidence; taking the multi-source dynamic fluctuation indicator and the plurality of fault hierarchical confidence pairs as joint retrieval conditions, traversing the physical contradiction evidence priority rule library, and calling a plurality of matching confidence resetting operators; after the confidence resetting operation of the plurality of fault hierarchical confidence pairs is performed by using the matching confidence resetting operator, the first set of fault hierarchical confidence is replaced and updated according to the resetting result, and the first set of reset hierarchical confidence is output.

5. The terminal temperature monitoring and fault automatic diagnosis method of the smart measurement switch according to claim 1, wherein, The state transition probability correction is performed on the P initial fault prediction results to obtain P-1 updated fault prediction results, the method comprising: extracting a first predicted fault mode from a first initial fault prediction result; extracting a second predicted fault mode from a second initial fault prediction result; taking the first predicted fault mode as a transition starting state, loading the second predicted fault mode into the fault state transition probability matrix, and querying a second transition probability weight; using the second transition probability weight to correct the initial confidence of the second initial fault prediction result to obtain a second updated fault prediction result; by analogy, the state transition probability correction is performed on the P initial fault prediction results to obtain the P-1 updated fault prediction results.

6. The terminal temperature monitoring and fault automatic diagnosis method of the smart metering switch according to claim 5, wherein, If the fault transition chain from the first predicted fault mode to the second predicted fault mode does not satisfy the fault legal transition path, the second updated fault prediction result is rolled back to the second initial fault prediction result.

7. The terminal temperature monitoring and fault automatic diagnosis method of the smart measurement switch according to claim 1, wherein, If the long-term and short-term temperature time-varying characteristics satisfy a preset temperature anomaly judgment condition, the start and end time stamps of the long-term and short-term temperature time-varying characteristics are called, comprising: decomposing the long-term and short-term temperature time-varying characteristics to obtain a short-term temperature change rate, a long-term temperature rise trend slope, and a neighboring terminal temperature standard deviation; using the temperature anomaly judgment condition to traverse the short-term temperature change rate, the long-term temperature rise trend slope, and the neighboring terminal temperature standard deviation, when any feature continuously satisfies the abnormal condition, backtracking to locate the first abnormal data sampling time; calling the last abnormal data sampling time from the long-term and short-term temperature time-varying characteristics, and the first abnormal data sampling time and the last abnormal data sampling time constitute the start and end time stamps.

8. The terminal temperature monitoring and fault automatic diagnosis method of the smart metering switch according to claim 7, wherein, The temperature anomaly judgment condition includes a transient thermal stress critical rate, an insulation degradation temperature rise gradient threshold, and a spatial thermal imbalance deviation limit.

9. The terminal temperature monitoring and fault automatic diagnosis method of the smart measurement switch according to claim 1, wherein, The method further comprises: performing local maintenance record calling of the intelligent measurement switch to obtain N fault maintenance data, wherein each fault maintenance data includes a fault maintenance difficulty coefficient and a fault maintenance effect coefficient, and N is a positive integer; based on the N fault maintenance data, performing fault maintenance quality evaluation to obtain a fault maintenance quality index; when the fault maintenance quality index is lower than a preset quality standard, performing replacement operation of the intelligent measurement switch.

Images