Power equipment operation and maintenance early warning method and system based on AI and Internet of Things big data

Abstract

The invention provides a power equipment operation and maintenance early warning method and system based on AI and Internet of Things big data. The method comprises the following steps: uniformly mapping power equipment operation data to a standardized feature space; an AI algorithm is used for learning boundary features of normal operation of equipment in a standardized feature space and an association rule between heterogeneous equipment, a self-adaptive adjustment mechanism of an early warning trigger threshold value is established, and a multi-dimensional condition combination is dynamically set; constructing a high-dimensional feature vector of a multi-dimensional parameter, and determining an accurate trigger boundary under different parameter combinations by using an AI algorithm; flexible matching and linkage triggering of early warning conditions among heterogeneous devices are achieved through a fuzzy inference rule base, and when a device operation boundary changes, the influence of the device operation boundary on the early warning conditions of the associated heterogeneous devices is automatically analyzed, and a multi-dimensional triggering condition combination is dynamically optimized. According to the invention, high efficiency, accuracy and intelligence of operation and maintenance of the power equipment can be realized, and the overall stability and early warning response capability of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment operation and maintenance, and particularly relates to a power equipment operation and maintenance early warning method and system based on AI and Internet of Things big data. BACKGROUND

[0002] In modern power systems, equipment operation and maintenance management is a crucial field that directly relates to the stability and safety of power supply. With the continuous expansion of the power grid scale and the increase in equipment complexity, how to ensure the reliability of equipment operation has become a core issue in the industry development. Power equipment operation and maintenance not only needs real-time monitoring of equipment status, but also needs timely early warning before potential failure occurs to avoid serious consequences caused by large-scale power outages or equipment damage.

[0003] However, there are still many challenges in the current power equipment operation and maintenance. Many existing methods often have difficulty in unified analysis when dealing with data from different sources, leading to information silos phenomenon, affecting the comprehensive judgment of equipment status. At the same time, the setting of early warning mechanism depends on fixed standards, which is difficult to adapt to the changes of equipment operation environment or the differences caused by equipment aging, which greatly reduces the accuracy and timeliness of early warning. The deeper problem is that the mutual influence between devices is often ignored, and the abnormality of a single device may trigger a chain reaction, but the existing methods lack effective evaluation of such correlation.

[0004] Therefore, the accurate evaluation of the correlation between devices has become a key factor to be solved. The correlation between devices refers to the fact that the state change of a device may indirectly affect other devices through operating parameters or environmental conditions, forming a complex interaction network. Due to the failure to effectively identify and quantify this correlation, there are often incomplete or false early warnings in the operation process. For example, in a substation, the abnormal temperature rise of a transformer may lead to an increase in the current load of adjacent devices, but if this correlated change cannot be perceived in advance, the best intervention opportunity may be missed, further exacerbating equipment wear and even causing system failure. The complexity of this correlation not only lies in the direct parameter transmission between devices, but also includes the comprehensive effect of environmental factors such as humidity and temperature changes on the device group, increasing the difficulty of problem judgment. How to accurately identify and quantify the correlation between devices in a complex operating environment and optimize the early warning mechanism based on this has become a key problem in the intelligent development of power equipment operation and maintenance. SUMMARY

[0005] The present application provides a power equipment operation and maintenance early warning method based on AI and Internet of Things big data, aiming to realize the efficiency, accuracy and intelligence of power equipment operation and maintenance, and improve the overall stability and early warning response capability of the system.

[0006] In a first aspect, the application provides a power equipment operation and maintenance early warning method based on AI and Internet of Things big data, mainly comprising: A heterogeneous data standardization processing framework is established to uniformly map power equipment operation data of different manufacturers and models to a standardized feature space; an AI algorithm is used to learn the boundary features of normal operation of the equipment in the standardized feature space and the correlation rules between heterogeneous equipment, an adaptive adjustment mechanism of the early warning trigger threshold is established, and multi-dimensional condition combinations are dynamically set; a high-dimensional feature vector of multi-dimensional parameters is constructed in the standardized feature space, an AI algorithm is used to learn the coupling relationship between parameters, and the accurate trigger boundary under different parameter combinations is determined; a fuzzy threshold boundary identification mechanism is established, a fuzzy set modeling is performed on the parameters in the high-dimensional feature vector, the accurate threshold is converted into a fuzzy interval with a membership function, and the width of the fuzzy boundary is dynamically adjusted according to the equipment data quality and historical reliability; the fuzzy reasoning rule library is used to realize flexible matching and linkage triggering of early warning conditions between heterogeneous equipment, when the operation boundary of a certain equipment changes, the influence of the change on the early warning conditions of associated heterogeneous equipment is automatically analyzed, and the multi-dimensional trigger condition combination is dynamically optimized.

[0007] Further, the establishment of the heterogeneous data standardization processing framework comprises: collecting temperature, current, and vibration parameter data of the power equipment under different seasons and different load conditions; uniformly mapping the temperature, current, and vibration parameter data to the standardized feature space; and using an AI algorithm to obtain the boundary features of normal operation of the equipment from the standardized feature space.

[0008] Further, the dynamic setting of the multi-dimensional condition combination comprises: determining a single parameter threshold trigger mode; determining a multi-parameter correlation trigger mode; calculating the correlation degree between parameters according to the multi-parameter correlation trigger mode to determine whether to trigger an early warning; determining a trend change rate trigger mode; and adjusting the trigger rules in the multi-dimensional condition combination according to the historical early warning accuracy.

[0009] Further, the construction of the high-dimensional feature vector of multi-dimensional parameters comprises: constructing transformer temperature, current, and oil level parameters into the high-dimensional feature vector; using an AI algorithm to learn the coupling relationship between parameters of the high-dimensional feature vector in historical operation data; determining the accurate trigger boundary for different seasons and load conditions; establishing an accuracy evaluation index system to monitor the determination accuracy of each parameter threshold; and dynamically adjusting the weight coefficients of the multi-parameter correlation trigger.

[0010] Further, the fuzzy set modeling of the parameters in the high-dimensional feature vector comprises: establishing membership functions for temperature and current parameters; converting the accurate threshold into the fuzzy interval; adjusting the width of the fuzzy boundary to a wider interval according to the equipment whose data accuracy is lower than a first preset threshold; adjusting the width of the fuzzy boundary to a narrower interval according to the equipment whose data accuracy is higher than a second preset threshold; and constructing the fuzzy inference rule base in the multi-dimensional parameter space.

[0011] Further, the flexible matching and linkage triggering of the early warning conditions between heterogeneous equipment are implemented, comprising: when the operation boundary of the transformer changes, the correlation law is obtained; the influence of the operation boundary change on the early warning condition of the switch cabinet is analyzed; if the influence exceeds a preset correlation degree, the triggering boundary of the power transmission line equipment is adjusted; the width of the fuzzy interval is updated synchronously; and the combination of the multi-dimensional triggering conditions is optimized.

[0012] Further, the weight coefficient of the multi-parameter correlation triggering is dynamically adjusted, comprising: for the case that the temperature parameter approaches the threshold but the current parameter is in a safe range, the correlation degree between the parameters is calculated; the weight coefficient is updated according to the accuracy evaluation index system; and the decision logic is adjusted.

[0013] Further, the fuzzy inference rule base in the multi-dimensional parameter space is constructed, comprising: determining an inference rule for a parameter combination in the fuzzy interval; calculating the membership degree of each parameter according to the membership function; and determining the early warning condition of the heterogeneous equipment through the fuzzy inference rule base.

[0014] In a second aspect, the power equipment operation and maintenance early warning system based on AI and Internet of Things big data comprises: A mapping module is configured to establish a heterogeneous data standardization processing framework, and map power equipment operation data of different manufacturers and models to a standardized feature space uniformly; A dynamic setting module is configured to learn the boundary features of equipment normal operation in the standardized feature space and the correlation law between heterogeneous equipment by using an AI algorithm, establish an adaptive adjustment mechanism of early warning triggering threshold, and dynamically set a multi-dimensional condition combination; A learning and determining module is configured to construct a high-dimensional feature vector of multi-dimensional parameters in the standardized feature space, learn the coupling relationship between parameters by using an AI algorithm, and determine accurate triggering boundaries under different parameter combinations; An establishing and adjusting module is configured to establish a fuzzy threshold boundary identification mechanism, model parameters in the high-dimensional feature vector as fuzzy sets, convert an accurate threshold into a fuzzy interval with a membership function, and dynamically adjust the width of the fuzzy boundary according to equipment data quality and historical reliability; The matching trigger module is used for realizing flexible matching and linkage triggering of early warning conditions between heterogeneous devices by using a fuzzy inference rule base, and when the operation boundary of a device changes, the influence of the device on the early warning conditions of associated heterogeneous devices is automatically analyzed and the multi-dimensional trigger condition combination is dynamically optimized.

[0015] Further, the mapping module comprises: The collection unit is configured to collect temperature, current and vibration parameter data of the power equipment under different seasons and different load conditions. The mapping unit is configured to map the temperature, current and vibration parameter data to the standardized feature space. The acquisition unit is configured to acquire the boundary features of the normal operation of the equipment from the standardized feature space by using an AI algorithm.

[0016] The technical scheme provided by the embodiment of the present application can include the following beneficial effects: The present application maps the equipment operation data of different sources to the standardized feature space, learns the normal operation boundary and associated rules of the equipment by combining an artificial intelligence algorithm, and establishes an adaptive early warning mechanism. The present application realizes accurate early warning and linkage adjustment of equipment parameters by dynamically optimizing multi-dimensional trigger conditions, combining fuzzy threshold boundary identification and flexible matching rules. Especially under complex conditions such as equipment aging and environmental changes, the present application ensures the accuracy of early warning by dynamically adjusting the fuzzy interval and the associated weight. At the same time, the present application uses high-dimensional feature vectors and a fuzzy inference rule base to analyze the influence propagation path between devices, synchronously optimizes the trigger boundary of related devices, and finally realizes the efficiency, accuracy and intelligence of power equipment operation and maintenance, and improves the overall stability and early warning response capability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The flowchart of the power equipment operation and maintenance early warning method based on AI and Internet of Things big data provided by the embodiment of the present application.

[0018] Figure 2 The module structure schematic diagram of the power equipment operation and maintenance early warning method based on AI and Internet of Things big data provided by the embodiment of the present application.

[0019] Figure 3 The module structure schematic diagram of the power equipment operation and maintenance early warning method based on AI and Internet of Things big data provided by the embodiment of the present application.

[0020] Figure 4 The functional module block diagram of the power equipment operation and maintenance early warning method based on AI and Internet of Things big data provided by the embodiment of the present application. DETAILED DESCRIPTION

[0021] For further understanding of the present application, the application will be described in detail in conjunction with the accompanying drawings and examples. The present application will be further described in detail in conjunction with the accompanying drawings and examples. It can be understood that the specific examples described herein are only for the purpose of explaining the related application, but not limiting the application. In addition, it should be noted that only the parts related to the application are shown in the drawings for the convenience of description.

[0022] As Figures 1-4 The power equipment operation and maintenance early warning method based on AI and Internet of Things big data provided by the embodiments of the application can specifically include the following steps: S100, a heterogeneous data standardization processing framework is established, and the operation data of power equipment of different manufacturers and models is uniformly mapped to a standardized feature space.

[0023] The operation data is obtained from power equipment of different manufacturers and models, the temperature, current and vibration parameters of transformers, switch cabinets and transmission lines and other equipment are preliminarily collected, and these raw data are converted into standardized data sets in a unified format according to the pre-established mapping rules, and stored in a shared data pool, laying a foundation for subsequent processing. For the standardized data sets in the shared data pool, a multi-dimensional feature space is constructed, and the temperature, current, vibration and other parameters are mapped into the space.

[0024] Step S200: learning the boundary features of the normal operation of the equipment in the standardized feature space and the correlation rules between heterogeneous equipment by using an AI algorithm, establishing an adaptive adjustment mechanism of the early warning trigger threshold, and dynamically setting a multi-dimensional condition combination.

[0025] The boundary features of the normal operation of the equipment are extracted by using a machine learning method, and a running boundary description set of each equipment type is generated, which is used for subsequent correlation rule mining. Based on the running boundary description set, the parameter correlation rules between heterogeneous equipment are analyzed, the linkage mode of the mutual influence between equipment is determined, and the correlation mapping table across equipment types is formed. When the running boundary of a certain equipment type changes, the affected other equipment types are identified through the mapping table, and the related early warning conditions are adjusted. For the early warning condition adjustment requirement in the correlation mapping table, combined with the historical early warning accuracy, the equipment aging degree and the running environment change data, the multi-dimensional trigger condition combination is dynamically updated, an adaptive early warning threshold configuration scheme is formed, and the accuracy and timeliness of the power equipment operation and maintenance early warning are ensured.

[0026] In one possible implementation, obtaining operation data from power equipment of different manufacturers and models involves a real-time monitoring process. For example, when collecting temperature parameters for a transformer, the data points are captured by Internet of Things sensors every fixed time interval, and the data is converted into a standardized data set after preliminary noise filtering. This method can ensure data consistency, facilitate subsequent unified processing, avoid heterogeneity-induced deviation, and thus improve overall early warning accuracy.

[0027] Specifically, for the current parameter acquisition of the switch cabinet, a wireless transmission module is used to obtain the original signal from the equipment end, and according to the pre-established mapping rules such as unit conversion and format alignment, it is stored in the shared data pool. This processing helps to integrate multi-source data, form a complete data set, and support cross-device analysis. Its beneficial effect is to reduce the data island phenomenon and improve the system response efficiency.

[0028] In a possible implementation, constructing a multi-dimensional feature space for the standardized data set in the shared data pool means creating an abstract coordinate system, where temperature corresponds to one dimension, current corresponds to another dimension, vibration corresponds to a third dimension, etc. The parameter values are projected into this space, and a machine learning method such as support vector machine is used to extract the boundary features of the normal operation of the equipment. Support vector machine is a supervised learning algorithm that classifies data by finding the maximum interval hyperplane. In this process, the standardized data set is input as a training sample, and the algorithm learns the boundary between normal and abnormal states to generate a running boundary description set. This extraction is beneficial to identifying the health pattern of the equipment and avoiding false judgments of fixed thresholds. Its effect is reflected in dynamically adapting to different operating scenarios and improving the robustness of early warning. For example, for the vibration parameters of the power transmission line, after mapping to the feature space, the support vector machine can analyze the historical data and extract the boundary such as the upper limit of the vibration amplitude, thereby providing a basis for correlation rule mining.

[0029] In a possible implementation, analyzing the parameter correlation rules between heterogeneous devices based on the running boundary description set involves correlation calculation, such as calculating the correlation coefficient of the transformer temperature and the switch cabinet current, determining the linkage mode through statistical methods, and forming a cross-device type correlation mapping table. This table records the influence path of the power transmission line vibration when the temperature of the transformer rises, and when the running boundary of a certain device type changes, such as the expansion of the temperature boundary, the affected other device types are identified through the mapping table, and then the related early warning conditions are adjusted. This analysis is beneficial to the realization of device group cooperation, and its effect is to timely capture the chain reaction and reduce the risk of the system. For example, in a high-temperature environment, the boundary change of the transformer can be linked to adjust the current early warning of the switch cabinet to prevent cascading failures.

[0030] In a possible implementation, the demand for early warning condition adjustment in the correlation mapping table is combined with the historical early warning accuracy such as the past false alarm rate, the device aging degree such as the running time data, and the running environment change such as seasonal load fluctuation, to dynamically update the multi-dimensional trigger condition combination, such as combining a single temperature threshold with a trend change rate to form an adaptive early warning threshold configuration scheme. This update is beneficial to optimizing the trigger rule, and its effect is reflected in reducing false negatives and improving operation and maintenance efficiency, supporting the accuracy and timeliness of the overall early warning of power equipment.

[0031] S300, constructing a high-dimensional feature vector of the multi-dimensional parameters in the standardized feature space, learning the coupling relationship between the parameters by using an AI algorithm, and determining the accurate triggering boundary under different parameter combinations.

[0032] The multi-dimensional parameter data of transformer temperature, current, oil level, etc. are obtained from the standardized feature space, and these parameters are constructed into a high-dimensional feature vector. The performance value of each group of vector data under different operating conditions is recorded to form an initial feature vector set. By using the feature vector set, the support vector machine algorithm is used to learn the coupling relationship between the multi-dimensional parameters, determine the mutual influence degree of each parameter under different combinations, generate a parameter coupling relationship table, and use it as the basis for subsequent boundary division. According to the parameter coupling relationship table, the contribution degree of each parameter in the high-dimensional feature vector is calculated to determine the accurate boundary range of the multi-parameter associated triggering, and the corresponding weight coefficient is allocated according to the contribution degree to form a mapping rule of boundary and weight. The mapping rule of boundary and weight is applied to the real-time operating data. When the multi-dimensional parameters approach the boundary range, the triggering condition is comprehensively judged according to the weight coefficient to complete the dynamic determination of the accurate boundary and the weight coefficient of the multi-parameter associated triggering.

[0033] In a possible implementation, multi-dimensional parameter data of transformer temperature, current, oil level, etc. are obtained from the standardized feature space, and these parameters are constructed into a high-dimensional feature vector. This construction process involves taking temperature data as the first dimension of the vector, current data as the second dimension, and oil level data as the third dimension. By integrating these parameters in the form of vectors, the internal relationship between the parameters can be effectively captured, which is beneficial to subsequent learning of parameter coupling relationship, because it can reduce the error of single parameter judgment and improve the accuracy of overall early warning.

[0034] Specifically, the performance value of each group of vector data under different operating conditions is recorded to form an initial feature vector set. For example, under the condition of high load in summer, the temperature may rise to a high level, while the current remains stable. This record helps to identify the pattern of parameter change, thereby providing a rich data basis for the AI algorithm and being beneficial to improving the learning efficiency and accuracy of the model.

[0035] In a possible implementation, the coupling relationship between the multi-dimensional parameters is learned by using the feature vector set and a support vector machine algorithm, which is a supervised learning method that classifies or regresses by finding the maximum interval hyperplane between data points. Here, the parameter coupling is learned, that is, how the current affects the oil level stability when the temperature rises. The process includes dividing the feature vector set into a training set and a test set, iteratively optimizing the hyperplane by the algorithm to minimize classification errors, and generating a parameter coupling relationship table, which is used as the basis for subsequent boundary division. For example, when the temperature and current are both high, the table records that their mutual influence degree is strong coupling, which is beneficial to avoid misjudgment caused by isolated parameter analysis and improve the adaptability of the system to complex operation scenarios.

[0036] In a possible implementation, for the parameter coupling relationship table, the contribution degree of each parameter in the high-dimensional feature vector is calculated. This calculation process evaluates the proportion of each parameter to the overall variation of the vector, for example, using principal component analysis to quantitatively determine that the contribution degree of the temperature parameter is higher than that of the oil level parameter. Then, the accurate boundary range of the multi-parameter association trigger is determined, and the corresponding weight coefficient is allocated according to the contribution degree, forming a mapping rule of the boundary and the weight. For example, the weight coefficient of the temperature is larger when the contribution degree is high. This is beneficial to dynamically adjust the trigger logic, so that the early warning is more in line with the actual device state, and the false positive alarm is reduced.

[0037] In a possible implementation, the mapping rule of the boundary and the weight is applied to the real-time operation data. When the multi-dimensional parameters approach the boundary range, the trigger condition is comprehensively judged according to the weight coefficient. This application process involves real-time comparison of the current vector and the mapping rule. If the temperature approaches the boundary but the total score does not exceed the threshold after weight adjustment, it is not triggered. The accurate boundary and the dynamic determination of the weight coefficient of the multi-parameter association trigger are completed, which is beneficial to realize efficient power equipment early warning and avoid the problem of insufficient accuracy caused by a single threshold.

[0038] S400, a fuzzy threshold boundary identification mechanism is established, the parameters in the high-dimensional feature vector are modeled by a fuzzy set, the accurate threshold is converted into a fuzzy interval with a membership function, and the width of the fuzzy boundary is dynamically adjusted according to the device data quality and historical reliability.

[0039] Raw operation data is obtained from power equipment of different manufacturers. For the temperature and current parameters of transformers and switch cabinets, an initial parameter set is constructed. These parameters are classified and organized according to equipment models and data acquisition frequencies to form a preliminary parameter distribution range, laying the foundation for subsequent processing. For the preliminary parameter distribution range, a fuzzy set model is constructed. The precise values of each parameter are converted into fuzzy intervals with membership functions. Through the pre-established membership mapping rules, the data of different equipment is mapped into a unified feature interval to form standardized fuzzy parameter intervals. On the basis of standardized fuzzy parameter intervals, the boundary width of the fuzzy interval is dynamically adjusted according to the historical operation records and data quality evaluation results of the equipment. For equipment with high data quality, the boundary range is narrowed, and for equipment with low data quality, the boundary range is expanded to form adaptively adjusted fuzzy boundary intervals. Using adaptively adjusted fuzzy boundary intervals, for the multi-dimensional combination of temperature and current parameters, a fuzzy reasoning rule library is preset to map the operating state of different equipment into a unified warning condition, realize the unified processing of heterogeneous data under the fuzzy threshold boundary recognition mechanism, and complete the fuzzy set modeling goal of equipment parameters.

[0040] For example, when processing data of power equipment from different manufacturers, raw operation data needs to be collected from transformers and switch cabinets first. Suppose the temperature data of a certain transformer is collected every minute, while the current data of a certain switch cabinet is collected every five minutes. These data differ in time frequency and dimension. By classifying these data according to equipment models and acquisition frequencies, a preliminary parameter distribution range can be formed, such as temperature data concentrated in a certain interval and current data distributed in another interval. The purpose of this is to ensure the pertinence and accuracy of data processing by subsequent unified processing according to different distribution characteristics.

[0041] Specifically, when constructing a fuzzy set model, precise temperature and current values can be converted into fuzzy intervals. Suppose the temperature reading of a certain transformer is a specific value, which can be mapped into a fuzzy interval, such as a high or low temperature state range, through a pre-set membership function. This conversion can better adapt to the differences between equipment, as the measurement accuracy of different equipment may result in slight deviations in values, and fuzzy intervals can accommodate these deviations, improving the fault tolerance of data processing. This processing method helps to avoid misjudgment caused by equipment differences in subsequent analysis.

[0042] In an embodiment, for the fuzzy parameter interval that has been formed, the dynamic adjustment of the boundary width can be made according to the historical operation records and data quality of the equipment. For example, the data acquisition device of an old switch cabinet has low precision, and its historical records show that the data fluctuates greatly, so a wider fuzzy boundary interval can be set for it to accommodate possible errors; while a newly installed transformer equipment has higher data quality, a narrower boundary interval can be set to improve the sensitivity of the judgment. The advantage of this is that it can be flexibly adjusted according to the actual situation of the equipment to ensure that the processing result is more in line with the actual operation state.

[0043] Then, using the adjusted fuzzy boundary interval, further processing can be carried out for the multi-dimensional combination of temperature and current. Assuming that the temperature of a transformer is in the high interval, and the current is also close to the upper limit interval, through the pre-set fuzzy reasoning rule library, this combination state can be mapped to the warning condition to judge the possible overload risk. This multi-dimensional combination processing method can comprehensively consider the correlation between multiple parameters to improve the comprehensiveness and reliability of the warning condition. In this way, heterogeneous data is effectively processed under the unified fuzzy threshold boundary recognition mechanism, fully adapting to the differences between different equipment.

[0044] S500, using the fuzzy reasoning rule library to realize flexible matching and linkage triggering of warning conditions between heterogeneous equipment, when the operation boundary of a certain equipment changes, automatically analyzing its influence on the warning conditions of associated heterogeneous equipment and dynamically optimizing the multi-dimensional trigger condition combination.

[0045] The operation data is obtained from power equipment of different manufacturers, and the temperature and current parameters of transformer, switch cabinet and other equipment are preliminarily sorted. The original data is mapped to the pre-established standardized feature space to form a unified feature data set, and the data source and historical reliability information of each equipment are retained to distinguish the data quality difference in subsequent processing. For the data quality difference in the standardized feature data set, a fuzzy set model is constructed to convert the accurate values of temperature and current parameters into fuzzy intervals with membership functions, and the width of the fuzzy interval is dynamically adjusted according to the historical reliability information of each equipment to form a fuzzy boundary set that adapts to the data precision of different equipment. On the basis of the fuzzy boundary set, a fuzzy reasoning rule library is pre-established to associate and map the fuzzy intervals of different equipment. When the operation boundary of a certain equipment changes, the influence range of its parameters on associated equipment is automatically identified to generate a corresponding warning condition adjustment scheme. According to the generated warning condition adjustment scheme, the linkage trigger conditions in the fuzzy reasoning rule library are updated to realize flexible matching of warning conditions between heterogeneous equipment, ensuring that when the operation boundary of a certain equipment changes, the warning conditions of associated equipment can be dynamically optimized to complete the linkage triggering of the warning conditions.

[0046] In a possible implementation, when obtaining operation data from power equipment of different manufacturers, the temperature parameter of the transformer equipment can be collected in Celsius unit, and the current parameter of the switch cabinet can be collected in ampere unit. The preliminary arrangement includes uniformly converting the original data to the same sampling frequency through linear interpolation method to form a unified feature data set, while retaining the data source such as manufacturer identification and historical reliability information such as equipment service life record. This can bring the beneficial effect of accurately distinguishing the data quality difference, because the low reliability of old equipment will lead to large data fluctuation. By retaining these information, subsequent processing strategy can be adjusted to ensure higher credibility of the overall data.

[0047] Exemplarily, when constructing a fuzzy set model for the data quality difference in the standardized feature data set, the accurate value of the temperature parameter, such as the reading of a certain transformer, is converted into a fuzzy interval of a triangular membership function, where the membership function represents the degree to which the data belongs to the normal range. The interval width is dynamically adjusted according to the historical reliability information, for example, the interval of a high-reliability device is narrow to plus or minus 1 unit, and the interval of a low-reliability device is wide to plus or minus 3 units, forming a fuzzy boundary set. This can bring the beneficial effect of adapting to different equipment data precision, because it allows low-precision equipment to have a larger tolerance range, avoiding frequent misjudgment due to precision difference, and improving the robustness of the early warning.

[0048] In a possible implementation, when pre-establishing a fuzzy reasoning rule library based on the fuzzy boundary set, the temperature fuzzy interval of the transformer and the current fuzzy interval of the switch cabinet are associated and mapped through IF-THEN rules, for example, if the transformer temperature interval exceeds the upper boundary, it is inferred that the switch cabinet current interval needs to be narrowed. When the operating boundary of a certain device such as a transformer changes, such as a sudden temperature rise, the influence range of the parameter of the associated device such as the switch cabinet is automatically identified, and the early warning condition adjustment scheme is generated by calculating the interval overlap degree. This can bring the beneficial effect of timely responding to equipment changes, because it ensures the quantitative evaluation of associated influence, and enhances the adaptability of the system to heterogeneous environment.

[0049] Exemplarily, when updating the linkage trigger condition in the fuzzy reasoning rule library according to the generated early warning condition adjustment scheme, the flexible matching of the early warning conditions between heterogeneous devices is realized, for example, when the transformer boundary change causes the scheme to indicate the switch cabinet threshold to be lowered, the rule is updated to flexibly match the new interval, ensuring the dynamic optimization of the early warning condition of the associated device and completing the linkage trigger. This can bring the beneficial effect of overall early warning accuracy, because it avoids the limitations of static threshold through continuous optimization, realizes real-time collaboration across device types, and maintains the stable operation and maintenance of power equipment.

[0050] The multi-dimensional trigger condition combination includes at least two modes of single parameter threshold triggering, multi-parameter correlation triggering, and trend change rate triggering.

[0051] Multi-dimensional parameters are obtained from power equipment operation data, including key indicators such as temperature, current, and oil level. When obtaining data, the parameter fluctuation characteristics under different operating environments and load conditions are considered, and a historical change record table of the parameters is constructed for subsequent determination of whether the parameters exceed the preset threshold range. For the parameter data in the historical change record table, a combination relationship matrix of multi-dimensional parameters is constructed to determine the boundary conditions of single parameter threshold triggering. At the same time, the potential conditions of multi-parameter correlation triggering are identified through correlation calculation between parameters, and the data in the combination relationship matrix are mapped into the early warning judgment logic. On the basis of the combination relationship matrix, the trend change rate of the parameters is calculated, and the determination conditions of the trend change rate triggering are determined for the short-term fluctuation range of indicators such as temperature or current. If the trend change rate exceeds the preset range, the single parameter threshold and the multi-parameter correlation condition are combined to form a comprehensive early warning logic. The comprehensive early warning logic is applied to the multi-dimensional trigger condition combination to ensure that at least two modes of single parameter threshold triggering, multi-parameter correlation triggering, and trend change rate triggering can work together, and the trigger conditions are dynamically adjusted for different equipment operating states to achieve accurate early warning judgment.

[0052] In one embodiment, the process of obtaining multi-dimensional parameters from power equipment operation data involves real-time collection of data streams of indicators such as temperature, current, and oil level. This acquisition method realizes continuous monitoring through Internet of Things sensors, thereby forming a complete data sequence that can capture the parameter change rules of the equipment under high-temperature summer or low-load winter. This helps to avoid the limitations of static thresholds in subsequent analysis and provides a more reliable early warning basis.

[0053] Specifically, when constructing the historical change record table of the parameters, the collected data is stored in a table structure according to the time stamp and environmental factors, for example, the temperature data is associated with the corresponding load condition to form a table structure, where each row represents a data point at a time point, and each column corresponds to a different parameter. This table can intuitively reflect the parameter fluctuation characteristics and improve the judgment efficiency, as it integrates historical context and avoids misjudgment caused by isolated data.

[0054] In one embodiment, the process of constructing a combination relationship matrix of multi-dimensional parameters for parameter data in the historical change record table includes calculating the correlation coefficients between parameters, for example, by statistical methods to obtain the positive correlation between temperature and current, thereby generating a matrix, where the matrix elements represent the coupling strength between parameters. This matrix construction helps to identify the boundaries of single parameter threshold triggering, for example, when the temperature exceeds the preset value, check whether the related elements in the matrix support triggering, thereby improving the accuracy of early warning and avoiding errors of single indicators.

[0055] Specifically, when identifying potential conditions triggering multi-parameter correlation, the correlation calculation in the matrix is utilized, such as integrating temperature and oil level data to assess comprehensive risks, which brings a more comprehensive equipment state assessment because it considers parameter interaction and avoids missing potential problems.

[0056] In an embodiment, the process of calculating the trend change rate of the parameter based on the combined relationship matrix involves a derivative approximation of the short-term data sequence, such as calculating the change slope of consecutive sampling points of the temperature index to quantify the fluctuation amplitude, which helps to determine the judgment condition of the trend change rate trigger because it captures dynamic anomalies rather than static values, improving the timeliness of the warning.

[0057] Specifically, if the trend change rate exceeds the preset range, the process of forming a comprehensive warning logic combining single parameter threshold and multi-parameter correlation conditions includes logical operations, such as using AND or OR gate rules to integrate multiple trigger signals, which brings a synergistic effect to the logic formation to ensure a more robust warning because it combines multiple modes and reduces the impact of environmental interference.

[0058] In an embodiment, the process of applying the comprehensive warning logic to the multi-dimensional trigger condition combination involves dynamically adjusting the weight, such as optimizing the priority of single parameter threshold trigger and trend change rate trigger according to the equipment state, which enables precise warning judgment because it ensures the synergistic effect of at least two modes, adapts to different operating states, and improves overall operation efficiency.

[0059] Specifically, this synergistic effect is reflected in practice when the temperature trend rises sharply, even if the single threshold is not reached, but combined with multi-parameter correlation, it can trigger, thereby preventing potential failures and bringing higher safety protection.

[0060] The dynamic optimization of multi-dimensional trigger condition combination includes continuously adjusting the trigger rules and personalizing the device aging factors according to historical operating data under different seasons and different load conditions.

[0061] The power equipment operation parameters under different seasons and different load conditions are obtained from historical operation data, and the temperature and current data of devices such as transformers and switch cabinets are classified and stored. These data are grouped according to time sequence and environmental conditions to form seasonal operation data sets and load variation data sets, which are used for subsequent preparation of rule adjustment basis data. For seasonal operation data sets and load variation data sets, multi-dimensional trigger rule initial templates are constructed, the fluctuation range of temperature and current is divided into multiple intervals, and the trigger condition combination of each interval is established in advance based on the distribution characteristics of historical data to generate a preliminary rule library, which provides a benchmark for subsequent dynamic adjustment. On the basis of the preliminary rule library, the aging degree related data of the device is obtained, the aging influence factor of each device is determined through comprehensive evaluation of the running time and maintenance records of the device, and the factor is embedded into the trigger condition combination to form a personalized rule adjustment scheme, which ensures that the rule library can adapt to the state difference of different devices. Starting from the personalized rule adjustment scheme, real-time data during device operation are continuously collected and compared with historical data. If the fluctuation of real-time data exceeds the preset threshold in the rule library, the trigger condition is fine-tuned, and the multi-dimensional trigger condition combination in the rule library is updated to maintain the dynamic optimization capability for different seasons and load conditions.

[0062] For example, when obtaining power equipment operation parameters under different seasons and different load conditions from historical operation data, the transformer temperature data is classified and stored, for example, the data of the summer high temperature period is grouped into a seasonal operation data set, which contains continuous time sequence temperature fluctuation records. This grouping can effectively capture the impact of seasonal changes on the device, thereby providing reliable basis data for subsequent rule adjustment, avoiding analysis bias caused by mixed data, and being beneficial to improving the accuracy of early warning.

[0063] In an embodiment, when constructing multi-dimensional trigger rule initial templates for seasonal operation data sets, the temperature fluctuation range is divided into three intervals, for example, the distribution characteristics of historical data are counted, and the middle interval is set as the normal operation boundary. This pre-established trigger condition combination generates a preliminary rule library, which can be used as a benchmark for dynamic adjustment, and is beneficial to reducing false positives of fixed thresholds. Through the combination with load variation data sets, the adaptability of the rules is further enhanced.

[0064] For example, when obtaining the aging degree related data of the device on the basis of the preliminary rule library, the aging influence factor is determined through comprehensive evaluation of the running time and maintenance records of the transformer, for example, the factor of the device running more than 10 years is set to a higher value, and is embedded into the trigger condition combination to form a personalized rule adjustment scheme. This embedding ensures that the rule library adapts to the state difference of the device, and is beneficial to relaxing the threshold for old devices to avoid false negatives, and lays a foundation for real-time data comparison.

[0065] In an embodiment, when real-time data is continuously collected and compared with historical data from the personalized rule adjustment scheme, if the real-time temperature fluctuation exceeds the preset threshold in the rule library, the trigger condition is fine-tuned, for example, the current threshold interval is adjusted under low-load winter conditions. This update can maintain the dynamic optimization capability of multi-dimensional trigger condition combinations for different seasons and load conditions, and is beneficial to the response efficiency and accuracy of the overall early warning system.

[0066] The fuzzy threshold boundary identification mechanism includes establishing membership functions for temperature, current, vibration and other parameters respectively, so that devices with lower data accuracy use wider fuzzy intervals, and devices with higher data accuracy use narrower fuzzy intervals.

[0067] According to the data acquisition characteristics of different devices, membership functions are constructed for temperature, current, vibration and other parameters, and the original collected data is mapped into fuzzy intervals. The width of the fuzzy interval is divided according to the historical data quality and measurement accuracy of the device, so that devices with lower data reliability correspond to wider fuzzy intervals, and devices with higher data reliability correspond to narrower fuzzy intervals. On the basis of the division of the fuzzy interval, the mapped temperature, current, vibration parameter data form a multi-dimensional fuzzy feature set, and the pre-established fuzzy reasoning rules are constructed for this set. The weight proportion of different parameters in reasoning is determined according to the device type and parameter characteristics, to ensure that the correlation between multi-dimensional features is reflected. For the matching process of the multi-dimensional fuzzy feature set and the fuzzy reasoning rule, the determination boundary value of each parameter on different devices is obtained, and the boundary width of the fuzzy interval is dynamically adjusted to ensure that flexible judgment can be made according to the accuracy difference of the device when the parameter fluctuates, forming a personalized fuzzy threshold range for each device. On the basis of the personalized fuzzy threshold range, the real-time collected data of temperature, current, vibration and other parameters are continuously compared. If the real-time data exceeds the fuzzy threshold range of the corresponding device, the early warning condition is triggered, to ensure that the fuzzy threshold boundary identification mechanism can adapt to the actual operating state of different accuracy devices.

[0068] For example, in the operation and maintenance of power equipment, according to the data acquisition characteristics of different devices, when constructing the membership function for the temperature parameter, the original collected data is mapped into the fuzzy interval through the triangular membership function. This function is defined as the membership degree is 1 when the data value is in the interval center, and gradually decreases to 0 towards both sides, so that the interval width is divided according to the historical data quality and measurement accuracy of the device, so that devices with lower data reliability correspond to wider fuzzy intervals, which can tolerate greater data fluctuations and avoid frequent misjudgments due to insufficient accuracy. Devices with higher data reliability correspond to narrower fuzzy intervals, improving the accuracy of the judgment. This can bring more reliable early warning triggering effect, because it adapts to the differences between devices and ensures the robustness of the overall system.

[0069] In a possible implementation, on the basis of the above-mentioned fuzzy interval division, the mapped temperature parameter data is combined with other parameter data such as current and vibration parameter data to form a multi-dimensional fuzzy feature set, and when the pre-established fuzzy reasoning rules are constructed for the set, the rules are defined by using the Mamdani fuzzy reasoning method. This method first fuzzifies the input, then applies the rule base for reasoning, and finally defuzzifies the output to obtain the output. The weight proportions of different parameters in reasoning are determined according to the device type and parameter characteristics, for example, the temperature weight is higher to highlight the thermal abnormal risk, and the relevance between multi-dimensional features is ensured to be reflected. In this way, a more comprehensive parameter linkage effect can be achieved, because it captures the coupling relationship between parameters, improves the accuracy of early warning, and provides a solid foundation for subsequent boundary adjustment.

[0070] Specifically, for the matching process of the above-mentioned multi-dimensional fuzzy feature set and fuzzy reasoning rules, when the determination boundary values of each parameter on different devices are obtained, the boundary width of the fuzzy interval is dynamically adjusted by calculating the average value of the membership degree. This adjustment process involves comparing the deviation of the current parameter value from the historical boundary value, and making a flexible judgment according to the device accuracy difference when the parameter fluctuates, for example, if the boundary of the vibration parameter fluctuates greatly, the interval is widened to form a personalized fuzzy threshold range for each device. In this way, more personalized operation and maintenance adaptability can be achieved, because it takes into account the specific conditions of the device, reduces the error caused by the universal threshold, and, combined with the aforementioned weight proportion, strengthens the continuity of the judgment.

[0071] In a possible implementation, on the basis of the above-mentioned personalized fuzzy threshold range, when the real-time collected data of parameters such as temperature are continuously compared, if the real-time data exceeds the fuzzy threshold range of the corresponding device, the early warning condition is triggered through logical judgment. This comparison process uses threshold comparison operation to ensure that the fuzzy threshold boundary recognition mechanism can adapt to the actual operating state of devices with different accuracies. In this way, timely and effective early warning response effects can be achieved, because it integrates the outputs of all the previous steps to realize a complete chain from data mapping to final determination, and improves the overall efficiency and safety of power equipment operation and maintenance.

[0072] The fuzzy reasoning rule base includes a rule set based on a multi-dimensional parameter space, which is used to determine whether to trigger early warning by calculating the correlation degree and membership degree when the parameter approaches the boundary.

[0073] The historical operation data in the multi-dimensional parameter space is collected to obtain real-time values and historical values of multiple parameters such as transformer temperature, current, oil level, etc., and each parameter is combined into a high-dimensional feature vector. A fuzzy set modeling is established based on the high-dimensional feature vector, and the accurate threshold of each parameter is converted into a fuzzy interval with a membership function, and the membership function represents the degree to which the parameter value belongs to the normal state or abnormal state. A fuzzy reasoning rule library is pre-established according to the fuzzy set modeling result, and the fuzzy reasoning rule library includes a rule set based on the multi-dimensional parameter space, and each rule is calculated by the correlation degree and the membership degree between parameters to determine whether to trigger a warning when the parameter is close to the boundary. When the parameter is close to the boundary, the rule set in the fuzzy reasoning rule library is called to calculate the membership degree of each parameter of the current high-dimensional feature vector, and fuzzy reasoning is performed in combination with the correlation degree between parameters to determine whether to trigger a warning.

[0074] Exemplarily, when collecting historical operation data, real-time values and historical values of parameters such as transformer temperature, current and oil level are obtained from Internet of Things power equipment, and these values are combined into a high-dimensional feature vector in a time series manner, which can capture the dynamic relationship between parameters, help to reduce the error of single parameter judgment in subsequent fuzzy modeling, and provide a more comprehensive warning basis.

[0075] In a possible implementation, the fuzzy set modeling is to convert the accurate threshold into a fuzzy interval, for example, setting a membership function for the temperature parameter, wherein the function value from 0 to 1 represents the gradual degree from abnormal to normal, which can handle data uncertainty, improve data compatibility between heterogeneous devices, and thus make the warning more flexible and accurate.

[0076] Specifically, the process of pre-establishing the fuzzy reasoning rule library involves creating a rule set according to the fuzzy set result, and each rule defines parameter correlation degree calculation, such as judging the overall state by combining the membership degree of current when the temperature is close to the threshold, which can realize coupled analysis of multi-dimensional parameters, avoid false positives, and enhance the adaptive ability of the system.

[0077] In a possible implementation, the rule set is called when the parameter is close to the boundary, the membership degree of the current high-dimensional feature vector is calculated and integrated into the correlation degree reasoning, for example, when the temperature rises but the oil level is stable under high load in summer, it is determined by the rule that no warning is triggered, which can optimize the triggering logic and improve the accuracy and response efficiency of the warning.

[0078] Exemplarily, from multiple aspects, this method uses a wide fuzzy interval to process low-precision data on old equipment, while a narrow interval is used to improve sensitivity on new equipment, and they mutually support to ensure the robustness of the overall system, which can balance the differences between different devices and realize a unified warning standard.

[0079] In a possible implementation, another direction is to adjust the rules for seasonal changes, such as strengthening the calculation of the association between the oil level and the temperature in winter, which can adapt to environmental drift and provide a continuously optimized early warning effect to support multi-scenario applications.

[0080] Specifically, these implementation methods form a complete chain through layer-by-layer logic from data collection to rule application, which can significantly reduce the risk of false negatives and improve the operation and maintenance reliability of power equipment.

[0081] The linkage trigger includes when a certain type of device parameter exceeds the fuzzy interval, the system automatically evaluates its impact on the early warning conditions of related heterogeneous devices and adjusts the trigger boundaries of other devices accordingly.

[0082] After obtaining the temperature parameter and the current parameter from the heterogeneous devices, the temperature parameter and the current parameter are mapped into a standardized feature space to obtain a standardized temperature value and a standardized current value. A membership function is established for the standardized temperature value and the standardized current value to determine the fuzzy interval boundary, and the width of the fuzzy interval boundary is dynamically adjusted according to the device data quality to form a first fuzzy interval and a second fuzzy interval. If the standardized temperature value of a certain type of device exceeds the first fuzzy interval, the influence degree is obtained by evaluating the associated influence of the exceeding on the standardized current value of the related heterogeneous devices through the pre-established fuzzy reasoning rule library. The trigger boundary of the second fuzzy interval of the related heterogeneous devices is adjusted according to the influence degree to realize the linkage trigger.

[0083] In a possible implementation, the temperature parameter and the current parameter are obtained from the heterogeneous devices such as transformers and switch cabinets. These parameters may be in different formats due to different manufacturers, such as transformer temperature recorded in Celsius and switch cabinet current in amperes. Through the mapping process, they are uniformly converted into values in the standardized feature space, which ensures the consistency of subsequent processing and helps to reduce the early warning deviation caused by data heterogeneity.

[0084] Specifically, the mapping involves dividing the original temperature parameter by a device-specific scaling factor to obtain a standardized temperature value, and similarly, the current parameter is also processed in the same way to become a standardized current value, thereby providing a unified basis for fuzzy modeling and avoiding errors caused by directly comparing original data. This uniformity can improve the accuracy of overall early warning because it eliminates the impact of device differences on evaluation.

[0085] In an embodiment, membership functions are established for the normalized temperature values and the normalized current values, defined as trapezoidal or triangular forms, where the membership from 0 to 1 represents the degree to which the parameter belongs to the normal range, and the fuzzy interval boundaries are dynamically adjusted in width according to the equipment data quality, for example, the boundaries of equipment with high data quality are narrow to increase sensitivity, and the boundaries of equipment with low data quality are wide to tolerate noise, so that the first fuzzy interval is formed for the temperature and the second fuzzy interval is formed for the current, and such adjustment can make the system more suitable for the actual operating environment and improve the robustness of the linkage triggering, because it considers the uncertainty caused by the reliability difference of the equipment.

[0086] For example, if the normalized temperature value of a certain type of equipment such as a transformer exceeds the first fuzzy interval, indicating a potential anomaly, an evaluation is made through a pre-established fuzzy reasoning rule base, which contains multiple if-then rules, such as if the temperature exceeds, then evaluate the impact on the current, and the evaluation process involves calculating the sum of the membership degrees of the exceeding part multiplied by the rule weight to obtain the impact degree, which quantifies the degree of abnormal propagation, helping to accurately capture the correlation between heterogeneous equipment and prevent small abnormalities from evolving into big problems.

[0087] Specifically, the construction of the fuzzy reasoning rule base is based on historical data learning rules, for example, using the Mamdani method to combine input fuzzy sets and defuzzify the output, which can handle uncertainty and improve the reliability of evaluation, because it simulates the application of expert knowledge in complex environments.

[0088] In a possible implementation, the second fuzzy interval trigger boundary of the related heterogeneous equipment is adjusted according to the obtained impact degree, for example, if the impact degree is high, then the current fuzzy interval of the switch cabinet is narrowed to reduce the trigger threshold, realizing linkage triggering, and such adjustment can respond to associated abnormalities in a timely manner and improve the early warning efficiency of the entire equipment group, because it ensures that abnormalities are captured before they spread, avoiding the omission caused by isolated processing.

[0089] For example, from multiple perspectives, such linkage can automatically tighten the current boundary of related lines when the transformer temperature is abnormal in high temperature seasons, preventing overload chain reactions, and relax the boundary to reduce false positives in low load periods, and these perspectives support each other, collectively enhancing the adaptability of early warning, because they all come from the unified logic of data mapping and fuzzy processing, ensuring stable operation and maintenance in heterogeneous environments.

[0090] Specifically, the impact degree serves as a bridge connecting the temperature and current interval adjustments, forming a closed loop from detection to response, and such closed loop can bring lower false positive rate and higher response speed, because it dynamically integrates the dependency relationship between equipment.

[0091] In an embodiment, an aging factor is incorporated into the impact degree calculation considering different device aging degrees, further refining the adjustment accuracy. This extension can make the linkage more personalized and improve the practicality of the overall system, as it adapts to the challenges brought by device life cycle changes.

[0092] The multi-parameter correlation trigger includes constructing transformer temperature, current, oil level and other parameters into a high-dimensional feature vector, and learning parameter drift characteristics under different operating conditions.

[0093] From historical operation data, obtain transformer temperature, current, oil level parameter sequences, and combine the parameter sequences into a high-dimensional feature vector. Cluster the distribution of the high-dimensional feature vector under different seasons and load conditions to divide multiple operating state subsets. Determine the boundary distribution of the high-dimensional feature vector within each operating state subset, and extract the trend of the boundary distribution changing with the operating state subset. Learn the parameter drift characteristics according to the trend of the boundary distribution, and adjust the decision boundary of the multi-parameter correlation trigger.

[0094] In a possible implementation, the process of obtaining transformer temperature, current, oil level parameter sequences from historical operation data involves extracting these parameters from time series data collected by Internet of Things devices, such as temperature sequences recorded as hourly reading sequences, current sequences recorded as peak sequences when load changes, and oil level sequences recorded as continuous sequences monitored by liquid level sensors. Combining the parameter sequences into a high-dimensional feature vector is done by constructing the temperature, current, and oil level values at each time point as vector components, such as a vector that includes a temperature component in Celsius, a current component in amperes, and an oil level component in percentage. This integrates multiple parameters into a unified representation, which is beneficial for capturing the mutual influence between parameters and thus improving the accuracy of early warning.

[0095] In a possible implementation, when clustering the distribution of the high-dimensional feature vector under different seasons and load conditions to divide multiple operating state subsets, first classify the vector distribution according to seasons such as summer high temperature period and winter low temperature period, and then subdivide according to load such as peak load and valley load. For example, the vector distribution under summer peak load may show that the temperature component is high and the current component fluctuates greatly. By using a clustering algorithm such as K-means clustering, similar distributed vectors are grouped into a subset. This division is beneficial for identifying operating modes under specific conditions, avoiding errors of single threshold judgment, and improving the adaptability of the system to environmental changes.

[0096] In a possible implementation, determining the boundary distribution of the high-dimensional feature vector within each subset of operating states involves calculating the statistical boundaries of the vectors in the subset, such as finding the upper and lower limits of the temperature component using the quantile method, and extracting the trend of the boundary distribution with the subset of operating states is achieved by comparing the differences in boundary values of different subsets, for example, the boundary shifts from high temperature to low temperature from the summer subset to the winter subset, which helps to quantify how the parameters drift with the conditions, supports more accurate early warning triggering, and reduces false positives.

[0097] In a possible implementation, learning the parameter drift characteristics according to the trend of the boundary distribution is to model the drift by analyzing the trend, such as the rising slope of the temperature boundary when the load increases, and adjusting the decision boundary of the multi-parameter correlation trigger is to update the triggering conditions accordingly, for example, when the trend shows that the oil level boundary decreases with the increase of current, the correlation threshold is correspondingly reduced, which helps to dynamically optimize the early warning logic and ensure accurate risk determination under different conditions, and achieve the accuracy of multi-parameter correlation triggering.

[0098] The weight coefficient adjustment includes establishing an accuracy evaluation index system, continuously monitoring the determination accuracy of each parameter threshold and updating the correlation weight accordingly.

[0099] From the historical operation data, the records of transformer temperature, current, oil level and other multi-dimensional parameters are obtained, an index system for evaluating the determination accuracy is constructed, the historical determination results of these parameters are compared with the actual fault records, and the accuracy score data of each parameter threshold determination is generated. For the accuracy score data, the determination bias of each parameter under different seasonal and load conditions is analyzed, the adjustment direction of the correlation weight between the temperature and current parameters is determined, and a weight update basis table reflecting the coupling relationship between the parameters is generated. According to the weight update basis table, the weight coefficients of the multi-parameter correlation trigger are adjusted, for the parameter combination with low accuracy score, the influence proportion in the determination is appropriately reduced, and the updated weight configuration scheme is generated. The updated weight configuration scheme is applied to the early warning determination logic, the correlation degree of each parameter in the multi-dimensional parameter space is recalculated to ensure that the determination logic is consistent with the accuracy score data, and the correlation weight adjustment result conforming to the current operating environment is formed.

[0100] The historical operation data is used to obtain records of multi-dimensional parameters such as transformer temperature, current, oil level, etc., and an index system for evaluating the accuracy is constructed, wherein the index system is composed of the accuracy rate defined as the proportion of correct triggering when actual failure occurs, the recall rate as the proportion of all actual failures correctly identified, and the F1 score as the harmonic mean of the two, and the historical determination results of these parameters are compared with the actual failure records, for example, the records of temperature exceeding the preset threshold but no failure occurring under the condition of high load in summer are compared with the actual failure cases, and the accuracy score data of each parameter threshold determination is generated, wherein the accuracy score data includes the F1 score value of each parameter combination to reflect the determination reliability.

[0101] For the accuracy score data, the determination deviation of each parameter under different seasons and load conditions is analyzed, for example, the low F1 score of the temperature parameter in winter under low load indicates that it is easily affected by the environment and deviates, so as to determine the correlation weight adjustment direction between the temperature and current parameters.

[0102] In one possible implementation, the coupling strength is quantified by calculating the Pearson correlation coefficient of the temperature deviation and the current stability, wherein the Pearson correlation coefficient is obtained by dividing the covariance of the temperature deviation sequence and the current sequence by the product of their standard deviations, and a weight update basis table reflecting the coupling relationship between parameters is generated, which lists the correlation coefficient of each parameter pair as the adjustment basis to ensure that the subsequent weights can compensate for the deviation.

[0103] According to the weight update basis table, the weight coefficient of the multi-parameter correlation trigger is adjusted, for example, for the temperature and oil level parameter combination with low correlation coefficient, the weight is reduced from the initial value to reduce the interference with the overall determination.

[0104] In one embodiment, the gradient descent method is used to iteratively optimize the weights, wherein the gradient descent method is used to minimize the deviation by calculating the partial derivative of the loss function with respect to the weights and updating the weights along the negative gradient direction, the influence proportion of the parameter combination with low accuracy score in the determination is appropriately reduced, and the updated weight configuration scheme is generated, which includes the adjusted weight value of each parameter to match the current deviation analysis result.

[0105] The updated weight configuration scheme is applied to the early warning determination logic, and the correlation degree of each parameter in the multi-dimensional parameter space is recalculated, for example, the temperature weight multiplied by its normalized value and the current weight multiplied by its normalized value are fused using weighted summation, to ensure that the determination logic is consistent with the accuracy score data, and the correlation weight adjustment result conforming to the current operating environment is formed, which is directly used to update the threshold determination accuracy and optimize the correlation weight accordingly, to achieve the goal of weight coefficient adjustment.

[0106] The automatic analysis impact includes calculating the change propagation path of the correlation law between devices in the standardized feature space by using an AI algorithm.

[0107] After mapping the heterogeneous power device parameters in the standardized feature space, the correlation law matrix between device parameters is obtained, which records the correlation coefficient between the transformer temperature and the switch cabinet current. For a change in the operation boundary of a device, the other device parameter path directly associated with the device is extracted from the correlation law matrix to form an initial change propagation path. The support vector machine is used to regress and fit the parameter sequence on the initial change propagation path to determine the propagation strength and direction of the change along the path. According to the propagation strength obtained by regression fitting, the warning threshold adjustment amount of each associated device on the path is calculated in turn to realize the calculation of the change propagation path of the correlation law between devices.

[0108] In a possible implementation, the standardized feature space is a process of uniformly converting parameters of different power devices such as transformers and switch cabinets into a comparable numerical range, for example, mapping the temperature from Celsius to the interval of 0 to 1, which facilitates subsequent calculation of the correlation law matrix, which is obtained by calculating the Pearson correlation coefficient between parameters, such as the correlation coefficient of 0.8 between the switch cabinet current when the transformer temperature rises, indicating a positive correlation. This matrix helps to identify the dependency between devices, thereby reducing false positives caused by isolated judgments in early warning and benefiting the overall system stability.

[0109] For example, when the operation boundary of a device such as a transformer changes, i.e., its temperature exceeds the normal range, a path is extracted from the correlation law matrix, such as a sequence from the transformer temperature to the switch cabinet current to the transmission line vibration, forming an initial change propagation path, which can capture how the change is transmitted from one device to another, helping to predict cascading failures in advance and avoid the evolution of a single device problem into a system-level interruption.

[0110] In a possible implementation, the support vector machine is a supervised learning algorithm used for regression fitting, which fits the parameter sequence data on the path by finding the best hyperplane, such as fitting the temperature-current-vibration sequence to determine the propagation strength such as the change amplitude decay of 0.5 and the direction such as the positive increase, which helps to quantify the degree of impact, thereby making the early warning more accurate and benefiting the optimization of resource allocation to reduce unnecessary maintenance.

[0111] For example, according to the fitted propagation strength, the warning threshold adjustment amount of each associated device is calculated, such as increasing the switch cabinet current threshold by 10% if the strength is 0.7, thereby realizing the calculation of the change propagation path, which is beneficial to dynamically adapt to the changes of the device group, improve the operation and maintenance efficiency, and reduce the risk.

[0112] In a possible implementation, the process of path calculation from matrix extraction to fitting to adjustment ensures that the correlation rules between heterogeneous devices are comprehensively analyzed, for example, in the high-load season, the transformer path change affects the threshold adjustment of the switch cabinet, which can prevent overload and help extend the service life of the device.

[0113] For example, from multiple perspectives, such as in a low-temperature environment, the path intensity direction may be reversed, and the fitting process needs to consider seasonal data, which is complementary to the correlation coefficient of the matrix, supports more robust early warning, and is beneficial to cope with environmental variations.

[0114] In a possible implementation, for old devices, the path is more conservative when the coefficient in the matrix is low, and more historical data is used for fitting to ensure that the adjustment amount is appropriate and is beneficial to compatibility with different device types.

[0115] For example, combined with trend changes, path calculation can be extended to multiple parameters, such as vibration added to the temperature-current path, and fitting determines the composite strength, which supports path accuracy from single to multi-dimensional, and is beneficial to comprehensive risk assessment.

[0116] The trigger boundary of the corresponding adjustment of other devices includes synchronously updating the fuzzy interval width and multi-parameter judgment logic of related devices.

[0117] After obtaining the temperature and current parameters from heterogeneous devices, the parameters are mapped to a standardized feature space to form unified mapping parameters. A fuzzy membership function is established for the unified mapping parameters to determine the fuzzy interval width of each device to form an initial fuzzy boundary. On the basis of the initial fuzzy boundary, the boundary change of at least one device is determined to determine the linkage influence between related heterogeneous devices to form linkage adjustment parameters. The fuzzy interval width and multi-parameter judgment logic of related devices are synchronously updated for the linkage adjustment parameters to realize the trigger boundary of the corresponding adjustment of other devices.

[0118] In a possible implementation, after obtaining the temperature and current parameters from heterogeneous devices, the process of mapping the parameters to a standardized feature space to form unified mapping parameters involves normalizing the original data of different manufacturer devices, for example, the temperature data provided by the transformer device may be in Celsius, and the current data of the switch cabinet may be in amperes. Through linear transformation, these data are converted into dimensionless standardized values, thereby eliminating the inconsistency caused by unit differences. This ensures the consistency of subsequent processing and helps improve the accuracy of early warning.

[0119] Specifically, the mapping process includes first collecting the original parameters of each device, and then applying the min-max normalization method to scale the parameter values to the range of 0 to 1 to form unified mapping parameters, which is beneficial to the comparison and correlation analysis between heterogeneous data.

[0120] In an embodiment, in the step of establishing fuzzy membership functions for unified mapping parameters, determining fuzzy interval widths of each device, and forming initial fuzzy boundaries, the fuzzy membership function refers to a mathematical description used to represent the degree to which a parameter value belongs to a normal range, for example, for a temperature parameter, the function can be defined in a trapezoidal form, where the core interval represents complete normality and the edge gradient represents uncertainty, so when determining the width, the width of the old device is wider to tolerate data noise, and the width of the new device is narrower to pursue accuracy, which can reduce false positives and adapt to device diversity.

[0121] It should be noted that, on the basis of the initial fuzzy boundaries, the process of determining the linkage influence between associated heterogeneous devices for at least one device running boundary change and forming linkage adjustment parameters is to identify changes by analyzing historical data, for example, when the running boundary of a transformer is adjusted due to seasonal changes, the system checks its current association with the switch cabinet, if the transformer temperature rises, it is often accompanied by fluctuations in the current of the switch cabinet, then generate adjustment parameters to reflect this linkage, which helps the overall system respond to environmental changes and improve the linkage of early warning.

[0122] In a possible implementation, in the implementation of synchronously updating the fuzzy interval width and multi-parameter judgment logic of related devices for linkage adjustment parameters, the multi-parameter judgment logic refers to a rule set combining temperature and current, for example, if the temperature exceeds the fuzzy interval and the current trend is rising, the alarm is triggered, the width is reduced or expanded according to the adjustment parameter, and the logic rule is modified to include new associations, which ensures flexible matching between heterogeneous devices and brings more reliable operation and maintenance warning effect.

[0123] Specifically, this synchronous update can support system stability from multiple aspects, for example, in the high temperature season, the transformer boundary change leads to the widening of the switch cabinet width, the trend check is added to the judgment logic, thereby avoiding single parameter misjudgment, from the perspective of device aging, if the boundary of a device is tightened due to aging, the logic of associated devices is strengthened accordingly Multi-parameter verification supports the robustness of overall early warning.

[0124] In an embodiment, for example, a fuzzy function is established after mapping the vibration parameters of a power transmission line, the width is adjusted according to reliability, when the line boundary change affects the transformer, the interval and logic of the latter are updated to form adjustment parameters, from the data quality side, the wide interval of low-precision devices reduces false negatives, and the narrow interval of high-precision devices improves response speed, these aspects support each other, forming a consistent heterogeneous data processing framework, emphasizing the beneficial effects of adaptability and accuracy.

[0125] The power equipment operation and maintenance early warning system based on AI and Internet of Things big data provided by the embodiment of the application mainly comprises: a mapping module 100, a dynamic setting module 200, a learning and determining module 300, an establishing and adjusting module 400 and a matching and triggering module 500. The mapping module 100 is used for establishing a heterogeneous data standardization processing framework, and mapping power equipment operation data of different manufacturers and models to a standardized feature space uniformly; The dynamic setting module 200 is used for learning boundary features of normal operation of equipment and association rules between heterogeneous equipment in the standardized feature space by using an AI algorithm, establishing an adaptive adjustment mechanism of early warning triggering thresholds, and dynamically setting multi-dimensional condition combinations; The learning and determining module 300 is used for constructing a high-dimensional feature vector of multi-dimensional parameters in the standardized feature space, learning coupling relationships between parameters by using an AI algorithm, and determining accurate triggering boundaries under different parameter combinations; The establishing and adjusting module 400 is used for establishing a fuzzy threshold boundary identification mechanism, modeling parameters in the high-dimensional feature vector in a fuzzy set, converting accurate thresholds into fuzzy intervals with membership functions, and dynamically adjusting the width of the fuzzy boundary according to equipment data quality and historical reliability; The matching and triggering module 500 is used for realizing flexible matching and linkage triggering of early warning conditions between heterogeneous equipment by using a fuzzy reasoning rule base, automatically analyzing the influence of a change in the operation boundary of a certain equipment on associated early warning conditions of heterogeneous equipment and dynamically optimizing the multi-dimensional triggering condition combination.

[0126] Further, the mapping module 100 comprises a collecting unit, a mapping unit and an acquiring unit. The collecting unit is used for collecting temperature, current and vibration parameter data of power equipment under different seasons and different load conditions; The mapping unit is used for mapping the temperature, current and vibration parameter data to the standardized feature space uniformly; The acquiring unit is used for acquiring boundary features of normal operation of the equipment from the standardized feature space by using an AI algorithm.

[0127] It should be noted that the modules or units provided by the embodiment of the application have the same implementation principles and generated technical effects as the foregoing method embodiments, and the specific working processes of the modules are the same as the corresponding processes in the foregoing method embodiments, which will not be described herein again.

[0128] The above embodiment is only one of the preferred embodiments of the present application, and should not be used to limit the protection scope of the present application, but any modification or polishing without substantial meaning made in the main design idea and spirit of the present application, and the technical problems solved are still consistent with the present application, and should be included in the protection scope of the present application.

Claims

1. A power equipment operation and maintenance early warning method based on AI and Internet of Things big data, characterized in that, The application relates to a power equipment early warning method and device. The application comprises: establishing a heterogeneous data standardization processing framework to uniformly map power equipment operation data of different manufacturers and models to a standardized feature space; using an AI algorithm to learn boundary features of normal operation of the equipment in the standardized feature space and correlation rules between heterogeneous equipment, establishing an adaptive adjustment mechanism of early warning trigger thresholds, and dynamically setting multi-dimensional condition combinations; constructing a high-dimensional feature vector of multi-dimensional parameters in the standardized feature space, using an AI algorithm to learn the coupling relationship between parameters, and determining accurate trigger boundaries under different parameter combinations; establishing a fuzzy threshold boundary identification mechanism, modeling parameters in the high-dimensional feature vector as a fuzzy set, converting accurate thresholds into fuzzy intervals with membership functions, and dynamically adjusting the width of the fuzzy boundary according to equipment data quality and historical reliability; 2. The method of claim 1, wherein, using a fuzzy reasoning rule base to realize flexible matching and linkage triggering of early warning conditions between heterogeneous equipment, automatically analyzing the influence of changes in the operation boundary of a certain equipment on the early warning conditions of associated heterogeneous equipment, and dynamically optimizing the multi-dimensional trigger condition combinations. The establishment of the heterogeneous data standardization processing framework comprises: collecting temperature, current and vibration parameter data of power equipment under different seasons and different load conditions; uniformly mapping the temperature, current and vibration parameter data to the standardized feature space; 3. The method of claim 1, wherein, using an AI algorithm to obtain boundary features of normal operation of the equipment from the standardized feature space. The dynamic setting of multi-dimensional condition combinations comprises: determining a single parameter threshold trigger mode; determining a multi-parameter correlation trigger mode; calculating the correlation degree between parameters according to the multi-parameter correlation trigger mode, and determining whether to trigger early warning; determining a trend change rate trigger mode; 4. The method of claim 1, wherein, adjusting the trigger rules in the multi-dimensional condition combinations according to the historical early warning accuracy. The construction of the high-dimensional feature vector of multi-dimensional parameters comprises: constructing transformer temperature, current and oil level parameters into the high-dimensional feature vector; using an AI algorithm to learn the coupling relationship between parameters in the high-dimensional feature vector in historical operation data; determining the accurate trigger boundaries for different seasons and load conditions; establishing an accuracy evaluation index system to monitor the determination accuracy of each parameter threshold; 5. The method of claim 1, wherein, dynamically adjusting the weight coefficients of the multi-parameter correlation trigger. The fuzzy set modeling of parameters in the high-dimensional feature vector comprises: establishing a membership function for temperature and current parameters; converting the accurate thresholds into the fuzzy intervals; adjusting the width of the fuzzy boundary to a wider interval according to equipment with data accuracy lower than a first preset threshold; adjusting the width of the fuzzy boundary to a narrower interval according to equipment with data accuracy higher than a second preset threshold; 6. The method of claim 1, wherein, constructing the fuzzy reasoning rule base in the multi-dimensional parameter space. The flexible matching and linkage triggering of early warning conditions between heterogeneous equipment comprises: when the operation boundary of a transformer changes, obtaining the correlation rules; analyzing the influence of the operation boundary change on the early warning conditions of a switch cabinet; if the influence exceeds a preset correlation degree, adjusting the trigger boundary of a power transmission line equipment; synchronously updating the width of the fuzzy interval; optimizing the multi-dimensional trigger condition combinations.

7. The method of claim 4, wherein, The dynamic adjustment of the weight coefficient triggered by the multi-parameter correlation includes: For the case where the temperature parameter approaches the threshold value but the current parameter is in the safe range, the correlation degree between the parameters is calculated; According to the accuracy evaluation index system, the weight coefficient is updated; Adjust the decision logic.

8. The method of claim 5, wherein, The construction of the fuzzy reasoning rule base in the multi-dimensional parameter space includes: Determine the inference rule for the parameter combination in the fuzzy interval; According to the membership function, the membership degree of each parameter is calculated; Determine the heterogeneous device early warning condition through the fuzzy reasoning rule base.

9. The power equipment operation and maintenance early warning system based on AI and Internet of Things big data, characterized in that, It includes: Establish a mapping module for establishing a heterogeneous data standardization processing framework to unify the operation data of power equipment of different manufacturers and models to a standardized feature space; Dynamic setting module, for learning the boundary characteristics of normal operation of equipment and the correlation law between heterogeneous devices in the standardized feature space by using AI algorithm, establishing a self-adaptive adjustment mechanism of early warning trigger threshold, and dynamically setting multi-dimensional condition combination; Learning and determining module, for constructing a high-dimensional feature vector of multi-dimensional parameters in the standardized feature space, learning the coupling relationship between parameters by using AI algorithm, and determining the accurate trigger boundary under different parameter combinations; Establish an adjustment module for establishing a fuzzy threshold boundary identification mechanism, model the parameters in the high-dimensional feature vector as a fuzzy set, convert the accurate threshold into a fuzzy interval with a membership function, and dynamically adjust the width of the fuzzy boundary according to the device data quality and historical reliability; The matching trigger module is used to realize the flexible matching and linkage triggering of the early warning condition between heterogeneous devices by using the fuzzy reasoning rule base. When the operation boundary of a device changes, the influence of the change on the early warning condition of the associated heterogeneous device is automatically analyzed and the multi-dimensional trigger condition combination is dynamically optimized.

10. The method of claim 1, wherein, The mapping module includes: A collection unit is used to collect temperature, current, and vibration parameter data of power equipment under different seasons and different load conditions; The mapping unit is used to map the temperature, current, and vibration parameter data to the standardized feature space; An acquisition unit is used to acquire the boundary characteristics of normal operation of the equipment from the standardized feature space by using AI algorithm.

Images