Power equipment production data intelligent analysis method and system

Abstract

The application provides a kind of electric power equipment production data intelligent analysis method and system, it is related to industrial production data processing technical field.The method includes: collecting the production observation data of electric power equipment and constructing normal sample set and defect sample set, obtaining multiple process reference sequences of electric power equipment, constructing multiple observation sequences and calculating the multiple shape difference characteristics of each observation sequence respectively, determining the starting difference and ending difference of each observation sequence in each production process, constructing the difference state adjacency sequence between any adjacent two production processes based on normal sample set, calculating the difference transmission parameter between any adjacent two production processes, extracting the difference transmission deviation feature of each abnormal sample, identifying multiple production abnormal transmission modes in defect sample set, extracting multiple difference transmission fingerprint features and used for real-time production detection of electric power equipment.The application improves the monitoring accuracy of electric power equipment production abnormalities.

Description

Technical Field

[0001] This invention relates to the field of industrial production data processing technology, specifically to an intelligent analysis method and system for power equipment production data. Background Technology

[0002] With the development of modern industry, the production process of power equipment has become increasingly complex. The production quality of various power equipment, such as high and low voltage switchgear, intelligent prefabricated substations, and outdoor modular substations, directly affects the safety and stability of the power system. To ensure that power equipment meets high quality standards during manufacturing, various automated production lines, data acquisition, and monitoring systems have been widely used.

[0003] In actual production, due to the diversity of equipment, the complexity of processes, and the interaction of multiple factors, accurate identification and location of anomalies in production process monitoring still face some challenges. Currently, many power equipment production data analysis methods monitor and analyze real-time data for each process. For example, they use predetermined thresholds or rules to determine whether the production status of a particular process meets standards. This approach, focusing on anomaly detection in a single process, ignores the correlation between processes and the transmission effect of differences. During production, small deviations or minor changes may be amplified or corrected in subsequent processes. Even small differences, accumulated throughout the production chain, can ultimately have a significant impact on the overall quality of the equipment.

[0004] Therefore, how to accurately capture the transmission of differences between processes and their impact on the quality of the final product through a more systematic analysis method has become an urgent technical problem to be solved in the production of power equipment. Summary of the Invention

[0005] This invention provides an intelligent analysis method and system for power equipment production data to address the problems existing in related technologies. The technical solution is as follows:

[0006] In a first aspect, embodiments of the present invention provide an intelligent analysis method for power equipment production data, comprising:

[0007] Collect production observation data of power equipment, classify the production observation data to generate multiple sets of normal samples and abnormal samples, and construct normal sample set and defect sample set of power equipment.

[0008] Obtain the process reference sequences corresponding to the power equipment in multiple production processes, construct the observation sequences of each normal sample and abnormal sample in each production process, and calculate multiple morphological difference features of each observation sequence based on the process reference sequences.

[0009] Multiple process difference sequences are constructed for each observation sequence based on multiple morphological difference characteristics. The starting and ending differences of each observation sequence in each production process are determined based on the process difference sequences.

[0010] By integrating multiple initial and final differences, a difference state adjacency sequence between any two adjacent production processes is constructed based on the normal sample set. The difference propagation parameter between any two adjacent production processes is calculated based on the multiple difference state adjacency sequences.

[0011] Based on multiple difference transmission parameters, calculate multiple difference contribution factors for each abnormal sample in the defect sample set, extract the difference transmission deviation features of each abnormal sample, identify multiple production abnormal transmission patterns in the defect sample set, extract multiple difference transmission fingerprint features, and use them for real-time production detection of power equipment.

[0012] Preferably, multiple process difference sequences are constructed for each observation sequence based on multiple morphological difference features. The determination of the initial and final differences for each observation sequence at each production process based on the process difference sequences includes:

[0013] The morphological difference feature is the distance value between the process reference sequence and the observation sequence under the observation window. This includes performing a sliding window process based on timestamp alignment on the process reference sequence and the observation sequence, generating a reference sequence segment and an observation sequence segment under each observation window, and calculating the distance value between the reference sequence segment and the observation sequence segment to obtain the morphological difference feature of the observation sequence.

[0014] Multiple morphological difference features of each production process are spliced ​​in time to construct the process difference sequence of each observation sequence in each production process. Based on the pre-set start state window and end state window, the sequence start segment and sequence end segment of each process difference sequence are extracted respectively. The start difference and end difference of each normal sample and abnormal sample are extracted from the sequence start segment and sequence end segment.

[0015] Preferably, multiple initial and final differences are merged to construct a difference state adjacency sequence between any two adjacent production processes. The difference propagation parameters between any two adjacent production processes are calculated based on these multiple difference state adjacency sequences, including:

[0016] Based on the production sequence of power equipment for multiple production processes, the starting and ending differences of multiple observation sequences in the normal sample set for each production process are combined to generate a difference propagation sample for each observation sequence.

[0017] Multiple difference transmission samples between two adjacent production processes are spliced ​​and fused to construct a difference state adjacency sequence between two adjacent production processes, including an end state sequence and an start state sequence. The correlation coefficient between the end state sequence and the start state sequence is calculated to obtain the difference transmission parameters between any two adjacent production processes in the normal sample set.

[0018] Preferably, based on multiple difference propagation parameters, multiple difference contribution factors are calculated for each abnormal sample in the defect sample set, and the difference propagation deviation characteristics of each abnormal sample are extracted to identify multiple production anomaly propagation patterns in the defect sample set, including:

[0019] Determine the initial and final differences for each outlier sample in each production process. Based on multiple difference propagation coefficients and final differences, calculate the unpropagated differences for each production process and record them as difference contribution factors.

[0020] Based on the initial and final differences, the observed transfer parameters between any two adjacent production processes in the abnormal sample are calculated. Combined with the difference transfer parameters between the two production processes, multiple difference transfer deviation characteristics of each abnormal sample are calculated.

[0021] By combining the difference contribution factor and the difference transmission deviation feature, an anomaly transmission sequence is constructed for each anomaly sample. Based on multiple anomaly transmission sequences, multiple groups of anomaly samples are clustered to generate multiple production anomaly transmission patterns in the defect sample set.

[0022] Preferably, extracting multiple differential fingerprint features and using them for real-time production inspection of power equipment includes:

[0023] Construct an abnormal transition path for each production anomaly transmission mode. The abnormal transition path includes multiple transition points. Based on the end differences of multiple abnormal samples in each production process and the multiple difference transmission deviation characteristics of each abnormal sample, calculate the transition intensity of each transition point. Construct a transition intensity sequence of the abnormal transition path based on the transition intensity of the transition points.

[0024] By combining the normal sample set, the abnormal intensity index of multiple abnormal samples in each production process in the production abnormality transmission pattern is calculated, and the abnormal intensity sequence of the production abnormality transmission pattern is generated. The transition intensity sequence and the abnormal intensity sequence are combined to generate the differential transmission fingerprint feature of the production abnormality transmission pattern.

[0025] After acquiring real-time production monitoring data of power equipment, the real-time difference features of the real-time production monitoring data in each production process are extracted, and a real-time process difference sequence of the real-time production monitoring data is constructed. The real-time transmission strength parameter of the real-time production monitoring data between any two adjacent production processes is calculated by combining multiple difference transmission parameters, and a real-time transmission strength sequence of the real-time production monitoring data is constructed.

[0026] Real-time production monitoring data is matched with multiple difference transmission fingerprint features based on real-time process difference sequences and real-time transmission intensity sequences to generate real-time production monitoring results for power equipment.

[0027] Preferably, the real-time production monitoring data is matched with multiple difference transmission fingerprint features based on the real-time process difference sequence and the real-time transmission intensity sequence to generate real-time production monitoring results for power equipment, including:

[0028] Calculate the abnormal intensity matching parameter between the real-time process difference sequence and the abnormal intensity sequence in the difference transmission fingerprint feature, calculate the abnormal transmission intensity matching parameter between the real-time transmission intensity sequence and the transition intensity sequence in the difference transmission fingerprint feature, fuse the abnormal intensity matching parameter and the abnormal transmission intensity matching parameter to obtain the abnormal pattern matching value between the real-time production monitoring data and multiple difference transmission fingerprint features, and generate the real-time production monitoring result of the power equipment based on the multiple abnormal pattern matching values.

[0029] Secondly, embodiments of the present invention provide an intelligent analysis system for power equipment production data, comprising:

[0030] The production data acquisition module is used to collect production observation data of power equipment, classify the production observation data to generate multiple sets of normal samples and abnormal samples, and construct normal sample sets and defect sample sets of power equipment.

[0031] The morphological difference analysis module is used to obtain the process reference sequences corresponding to multiple production processes of power equipment, construct the observation sequence of each normal sample and abnormal sample in each production process, and calculate multiple morphological difference features of each observation sequence based on the process reference sequence.

[0032] The difference extraction module is used to construct multiple process difference sequences for each observation sequence based on multiple morphological difference features, and to determine the start and end differences of each observation sequence in each production process based on the process difference sequences.

[0033] The difference propagation analysis module is used to integrate multiple initial and final differences, construct a difference state adjacency sequence between any two adjacent production processes based on the normal sample set, and calculate the difference propagation parameters between any two adjacent production processes based on multiple difference state adjacency sequences.

[0034] The anomaly propagation pattern recognition module is used to calculate multiple difference contribution factors for each anomaly sample in the defect sample set based on multiple difference propagation parameters, extract the difference propagation deviation features of each anomaly sample, identify multiple production anomaly propagation patterns in the defect sample set, extract multiple difference propagation fingerprint features, and use them for real-time production detection of power equipment.

[0035] The present invention has the following beneficial effects:

[0036] This invention quantifies the deviation transmission effect between processes in power equipment production by constructing a multi-process difference transmission analysis framework. First, it extracts the start / end differences based on the morphological difference features of the process reference sequence and the observation sequence, and calculates the difference transmission parameters between processes by fusing the adjacent sequences of the difference states of normal samples. Then, it identifies the abnormal transmission patterns of production by combining the difference contribution factors of abnormal samples with the transmission deviation features, extracts and fuses the difference transmission fingerprint features of transition intensity and abnormal intensity, and dynamically tracks the transmission path and cumulative effect of differences between processes. Before minor deviations cause obvious defects, it can accurately identify abnormal transmission patterns. Based on the real-time detection mechanism of fingerprint feature matching, it significantly improves the early perception capability of progressive quality risks, avoids the early warning lag caused by the lack of process correlation, and provides highly sensitive quality risk prevention and control support for power equipment production. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating an intelligent analysis method for power equipment production data provided in an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the structure of an intelligent analysis system for power equipment production data provided in an embodiment of the present invention. Detailed Implementation

[0039] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0040] like Figure 1 As shown in the figure, an intelligent analysis method for power equipment production data provided by an embodiment of the present invention includes the following:

[0041] Step S1: Collect production observation data of power equipment, classify the production observation data to generate multiple sets of normal samples and abnormal samples, and construct normal sample set and defect sample set of power equipment.

[0042] Specifically, IoT sensor devices installed on power equipment production lines collect various production observation data of the power equipment. For the production of different power equipment, such as distribution cabinets, substations, transformers, and circuit breakers, key parameters corresponding to the production process can be collected, such as current, voltage, temperature, pressure, torque, stroke, and speed. For example, in the production process of high-voltage circuit breakers, information such as contact pressure curves, spring compression speed curves, and screw torque change curves are collected during the arc-extinguishing chamber assembly process, and information such as plating deposition rate curves, current density fluctuation curves, and plating solution temperature change curves are collected during the contact silver plating process. This data is used to reflect the execution status and actual performance of each process. The collected data can be transmitted to an industrial cloud platform in real time and preliminarily processed and analyzed through edge computing nodes to ensure low-latency processing and real-time response.

[0043] For the collected production observation data, based on the final product quality inspection results corresponding to these samples, the production samples that meet the quality standards and have no obvious defects can be selected, marked as normal samples and a normal sample set can be constructed. The remaining production samples with production defects, such as those with quality abnormalities due to minor process fluctuations or operational errors, can be marked as abnormal samples and an abnormal sample set can be constructed.

[0044] Step S2: Obtain the process reference sequences corresponding to the power equipment in multiple production processes, construct the observation sequences of each normal sample and abnormal sample in each production process, and calculate multiple morphological difference features of each observation sequence based on the process reference sequences.

[0045] Specifically, in this stage, a process reference sequence for each production step is first obtained. This sequence represents the performance of the step under ideal or standard conditions and is a benchmark sequence corresponding to the normal production process, including the reference changes of different parameters during production. Those skilled in the art will understand that these reference sequences are typically determined based on historical data from a large number of normal samples or industry standards. This embodiment does not specifically limit the definition of the benchmark sequence.

[0046] For each production process in the power equipment manufacturing process, multiple process observation sequences are generated using collected production observation data, which are the time-series changes of different production parameters in each process. Each process observation sequence is compared and analyzed with a process reference sequence to calculate multiple morphological difference features corresponding to different observation sequences.

[0047] The morphological difference feature can be the morphological distance value between the process reference sequence and the observation sequence under the observation window. For the extraction process of morphological difference feature, the process reference sequence and the observation sequence can be timestamped and time-synchronized by sliding window processing. The corresponding reference sequence segment and observation sequence segment under the observation window with the same time interval can be extracted, and the morphological distance value between the reference sequence segment and the observation sequence segment can be calculated.

[0048] First, first-order difference detection is performed on both the reference sequence segment and the observed sequence segment. This involves calculating the difference between the current feature value and the previous feature value in the sequence as the first-order difference feature of the current feature value. This is used to construct a reference difference sequence for the reference sequence segment and an observed difference sequence for the observed sequence segment. Feature values ​​whose first-order difference features are greater than a preset mutation threshold are recorded as mutation points. Multiple mutation points are identified in both the reference and observed sequence segments, and a reference mutation feature sequence Bx={bx1,…,bxp} and an observed mutation feature sequence Br={br1,…,brq} are constructed, where p and q represent the total number of mutation points in the reference and observed sequence segments, respectively.

[0049] The morphological distance between the reference sequence fragment and the observed sequence fragment is calculated based on the reference mutation feature sequence and the observed mutation feature sequence. This distance serves as the morphological difference feature corresponding to the observed sequence, and multiple morphological difference features corresponding to different observed sequences are calculated accordingly. The calculation method for the morphological distance value is as follows: In the formula, dm is the morphological distance between the reference sequence fragment and the observed sequence fragment, bri is the i-th mutation point in the observed sequence fragment, and bxj is the j-th mutation point in the reference sequence fragment.

[0050] Step S3: Construct multiple process difference sequences for each observation sequence based on multiple morphological difference features, and determine the starting and ending differences of each observation sequence in each production process based on the process difference sequences.

[0051] Specifically, in each production process, each observation sequence is extracted from multiple observation windows to describe the morphological difference features that describe the difference between the parameters and the baseline information. The extracted morphological difference features are then spliced ​​together in chronological order to construct a continuous time series corresponding to each observation sequence in each production process, forming a process difference sequence that describes the difference changes of the process at different time points.

[0052] It is worth noting that the initial difference and the final difference are used to analyze the potential impact of the difference in process j on process j+1. This is because the difference in process j may be corrected or absorbed by process j+1, or it may be amplified or even have no impact on subsequent processes. In order to quantify the final difference in process j, the degree of impact of the difference at the beginning of process j+1 is extracted. The initial difference and the final difference represent the difference that exists at the beginning of the process and the difference that exists at the end of the process, respectively.

[0053] For the extraction of initial and final differences, after constructing the process difference sequence for each production step, considering the potential fluctuations at the beginning of production and the possibility of noise at the end, the initial and final segments of each process difference sequence are extracted based on pre-defined initial and final state windows. Specifically, the segment within a certain timeframe after the start of the process is taken as the initial segment, and the segment within a certain timeframe before the end of the process is taken as the final segment. The mean or median of multiple morphological difference features within these segments represents the initial / final difference of the production step. The initial difference reflects whether the equipment has reached its predetermined working state at the start of the process, while the final difference represents whether there is a deviation in the equipment state after the process is completed. By processing each observation sequence in this way, the initial and final differences corresponding to each normal and abnormal sample are finally obtained.

[0054] Step S4: Merge multiple initial and final differences, construct a difference state adjacency sequence between any two adjacent production processes based on the normal sample set, and calculate the difference propagation parameter between any two adjacent production processes based on the multiple difference state adjacency sequences.

[0055] Specifically, through the analysis of multiple normal samples, based on the initial and final differences of each process, a difference state adjacency sequence between any two adjacent production processes is calculated. This difference state adjacency sequence describes the transmission of difference state information between two adjacent production processes during the production of power equipment. Based on this, and using multiple difference state adjacency sequences, a difference transmission parameter is calculated for each pair of adjacent processes, describing how the final difference state of one production process affects the initial difference state of the next production process.

[0056] For the calculation of the difference propagation parameter, since multiple production processes of power equipment are often carried out in a specific sequence, and the ending state of each process is closely related to the starting state of the next process, the starting and ending differences of each observation sequence in the normal sample set for each production process are combined according to the production sequence of the power equipment. This process mainly focuses on two adjacent production processes, including accurately combining the ending difference of the current production process with the starting difference of the next production process to obtain one difference propagation sample of the observation sequence.

[0057] For each normal sample, the starting and ending differences of every two adjacent processes are combined sequentially according to the production sequence. The combined data is used to represent the difference propagation status between these two processes, i.e., the process of difference propagation between processes. In this way, a set of difference propagation samples can be constructed to represent the difference status between each pair of adjacent processes.

[0058] The difference transmission samples between each pair of adjacent processes are spliced ​​and fused to form a difference state adjacency sequence between each pair of processes. This sequence consists of two parts: one is the end state sequence representing the end state of the previous process, and the other is the start state sequence representing the start state of the subsequent process. The two state sequences form the difference transmission state information between adjacent processes.

[0059] For each pair of adjacent sequences of differing states, the difference propagation parameters are calculated based on the ending state sequence and the starting state sequence. In this process, the ending state sequence and the starting state sequence are first centered. Based on the centered ending state sequence and the starting state sequence, the direction coordination parameter and the amplitude coordination parameter of the adjacent sequences of differing states are calculated. The centered process includes calculating the sequence mean of the ending state sequence and the starting state sequence, and calculating the difference between the ending difference / starting difference and the corresponding sequence mean, thus obtaining the centered ending state sequence and the starting state sequence.

[0060] For the calculation of the directional coordination parameter, the end state sequence after centering is multiplied element by element with the start state sequence, that is, the product of the end difference and the start difference of the same normal sample is calculated. If the product is greater than 0, a same-direction count is performed to obtain the number of same-direction samples in the adjacent sequence of the difference state. The ratio of the number of same-direction samples to the total number of normal samples is calculated as the directional coordination parameter.

[0061] For the calculation of the amplitude coordination parameter, the amplitude difference value is calculated element-by-element between the centered final state sequence and the initial state sequence, where: In the formula, Let be the amplitude difference value of the k-th normal sample. , ε represents the final difference and initial difference after centering the k-th normal sample, respectively, and ε is a constant term.

[0062] The mean of the amplitude coordination parameters from multiple normal samples is calculated to obtain the amplitude coordination parameter. Finally, the product of the directional coordination parameter and the amplitude coordination parameter is calculated as the difference propagation parameter between two adjacent production processes. This parameter comprehensively describes the directional and amplitude propagation characteristics between the two processes. Specifically, it determines whether a large difference at the end of the current process indicates a large difference at the beginning of the next process, and whether the strength of the propagation is stable (e.g., maintaining a certain proportion or fluctuating). This better quantifies the propagation characteristics between the two processes, revealing the extent to which the ending difference of one process influences the beginning difference of the next process.

[0063] The above method allows us to calculate the difference propagation parameter between any two adjacent production processes in the multiple production processes of power equipment, within the information represented by a normal sample set. The difference propagation parameter for each production process reveals the magnitude of its influence on subsequent processes. A high propagation parameter indicates that the difference in the preceding process may significantly affect the production status of subsequent processes, thus requiring special attention; while a low propagation parameter indicates that the difference in that process has a relatively small impact on subsequent processes.

[0064] Step S5: Calculate multiple difference contribution factors for each abnormal sample in the defect sample set based on multiple difference transmission parameters, extract the difference transmission deviation features of each abnormal sample, identify multiple production abnormal transmission patterns in the defect sample set, extract multiple difference transmission fingerprint features and use them for real-time production detection of power equipment.

[0065] Specifically, based on the aforementioned difference propagation parameters, a difference contribution factor is calculated for each defect sample to represent the magnitude of the overall difference contribution of a certain process to the defect sample. Furthermore, difference propagation deviation features are extracted for each abnormal sample to identify multiple production anomaly propagation patterns within the defect sample set, providing data support for real-time production inspection.

[0066] For the calculation of the difference contribution factor, based on the initial and final differences of each extracted outlier sample in each production process, and combined with multiple difference propagation coefficients, the theoretical portion of the final difference of each production process that is not propagated to the next production process is calculated, resulting in the unpropagated difference for each production process. The calculated unpropagated difference is recorded as the difference contribution factor corresponding to the production process. Wherein, unpropagated difference = normalized final difference × (1 - difference propagation coefficient). For normalized final differences, the mean and standard deviation of the final differences for multiple final differences in the production process are calculated. For any normalized final difference in a production process, normalized final difference = (final difference - mean final difference) / standard deviation of final difference.

[0067] Furthermore, by combining the initial and final differences, the observed transfer parameter between any two adjacent production processes in the outlier sample is calculated. Specifically, this is the ratio of the initial difference of the next process to the final difference of the previous process, representing the actual difference transfer relationship exhibited by the sample data in the actual observation data. After calculating the observed transfer parameter between any two adjacent production processes for each outlier sample, the ratio of the observed transfer parameter to the difference transfer parameter between the two processes is denoted as the difference transfer deviation characteristic of the outlier sample between the two adjacent production processes, in conjunction with the difference transfer parameter between the two processes.

[0068] By combining multiple difference contribution factors and difference transmission deviation features of each abnormal sample, an abnormal transmission sequence is constructed for each abnormal sample. This sequence, composed of multiple difference contribution factors and difference transmission deviation features, contains rich characteristic information about the difference transmission of abnormal samples during the production process. It can reflect the difference changes of each process in the production process within different abnormal samples, revealing the difference transmission mode and abnormality pattern between different processes under abnormal production conditions. Based on the abnormal transmission sequences of multiple abnormal samples, clustering algorithms such as K-means and DBSCAN can be used to perform cluster analysis on these sequences. Through cluster analysis, similar abnormal samples are grouped into one category, forming different production abnormality transmission patterns. These patterns can reveal the difference transmission of different defect types in the production process. For example, some defect types may have special patterns or deviations in the difference transmission process of specific processes. These patterns can be used to identify and locate possible defect sources. By analyzing the transmission patterns of multiple abnormal samples, effective production abnormality transmission fingerprint features can be extracted for the detection and analysis of abnormalities in the real-time production process of power equipment, achieving early warning and quality prediction.

[0069] For the multiple production anomaly transmission patterns constructed above, the differential transmission fingerprint features of each production anomaly transmission pattern are extracted, and real-time production detection of power equipment is performed based on multiple differential transmission fingerprint features.

[0070] The extraction of difference propagation fingerprint features involves constructing anomaly transition paths for each production anomaly propagation pattern. Each anomaly transition path includes multiple transition points, specifically, two adjacent production processes constitute one transition point. Based on the multiple termination differences and difference propagation deviation features of each anomaly sample, the transition strength of each transition point is calculated. Specifically: In the formula, L represents the transition intensity of the transition point, E represents the termination difference involved in the transition point, and G represents the difference propagation deviation characteristic involved in the transition point. A transition intensity sequence of the production anomaly propagation mode is constructed based on the transition intensities of multiple transition points, reflecting the changing characteristics of difference propagation among multiple anomaly samples throughout the production process.

[0071] This study calculates the anomaly intensity index of multiple anomalous samples in each production process within a production anomaly transmission pattern, combining the normal sample set with the anomaly data. Specifically, it determines the termination difference of each anomalous sample in each production process from the defect sample set, calculates the mean of the termination differences of multiple normal samples in each production process, and corrects the termination difference of each anomalous sample in the corresponding production process based on this mean, obtaining the local difference intensity of each anomalous sample in each production process. The mean of the local difference intensity of multiple anomalous samples in each production process within the production anomaly transmission pattern is then calculated as the anomaly intensity index of the production anomaly transmission pattern in different production processes. Finally, an anomaly intensity sequence of the production anomaly transmission pattern is generated based on multiple anomaly intensity indices, reflecting the comprehensive anomaly intensity of multiple anomalous samples in different processes within the production anomaly transmission pattern. The transition intensity sequence and the anomaly intensity sequence are combined to generate the difference transmission fingerprint feature of the production anomaly transmission pattern, representing the local anomaly intensity distribution in different production processes under a specific production anomaly transmission pattern, as well as the transmission relationship of anomaly intensity between production processes.

[0072] For the process of real-time production monitoring of power equipment based on multiple difference transmission fingerprint features, after acquiring the real-time production monitoring data of the power equipment, the real-time difference features of the real-time production monitoring data for each production process are extracted in the same way as the local difference intensity calculation method described above, and a real-time process difference sequence of the real-time production monitoring data is constructed. Furthermore, using the same method as the aforementioned difference transmission deviation feature extraction process, the real-time transmission intensity parameter of the real-time production monitoring data between any two adjacent production processes is calculated based on multiple difference transmission parameters, and a real-time transmission intensity sequence of the real-time production monitoring data is constructed. The real-time process difference sequence and the real-time transmission intensity sequence constitute the production status features of the real-time production monitoring data.

[0073] Finally, based on the real-time process difference sequence and the real-time transmission intensity sequence, the real-time production monitoring data is matched with multiple difference transmission fingerprint features. This includes calculating the abnormal intensity matching parameter between the real-time process difference sequence and the abnormal intensity sequence in the difference transmission fingerprint features, as well as calculating the abnormal transmission intensity matching parameter between the real-time transmission intensity sequence and the transition intensity sequence in the difference transmission fingerprint features.

[0074] The calculation process for the anomaly propagation strength matching parameter involves sorting the intensity of the real-time propagation strength sequence and the transition intensity sequence in the differential propagation fingerprint feature, determining the index positions of the first r intensity values ​​of both, constructing a set of index positions for these intensity values ​​for both the real-time propagation strength sequence and the transition intensity sequence, and calculating the anomaly propagation strength matching parameter between the real-time propagation strength sequence and the transition intensity sequence in the differential propagation fingerprint feature as follows: In the formula, U is the anomaly transmission strength matching parameter. To transmit the set of index positions of the intensity sequence in real time, This represents the set of index positions for the transition intensity sequence. By quantizing the matching status between the window positions containing multiple key intensity values, the abnormal transmission intensity matching parameters between the real-time transmission intensity sequence and the transition intensity sequence are calculated.

[0075] After calculating the anomaly intensity matching parameter and the anomaly propagation intensity matching parameter, the two are fused to calculate the anomaly pattern matching value between real-time production monitoring data and multiple differential propagation fingerprint features. The anomaly pattern matching value is calculated as follows: Anomaly Pattern Matching Value = λ1 × Anomaly Intensity Matching Parameter + λ2 × Anomaly Propagation Intensity Matching Parameter, where λ1 and λ2 are weighting coefficients that can be reasonably set based on the experience and knowledge of production management technicians to meet the actual production anomaly monitoring accuracy requirements.

[0076] If the abnormal pattern matching value exceeds the preset abnormal threshold, it indicates a potential anomaly in the power equipment production process, triggering a timely alarm. This allows production managers to adjust the production process in real time, ensuring stable product quality. Those skilled in the art can further optimize the difference transmission fingerprint features by continuously accumulating real-time monitoring data, making production monitoring more accurate.

[0077] like Figure 2 As shown in the figure, an intelligent analysis system for power equipment production data provided in this embodiment of the invention includes:

[0078] The production data acquisition module is used to collect production observation data of power equipment, classify the production observation data to generate multiple sets of normal samples and abnormal samples, and construct normal sample sets and defect sample sets of power equipment.

[0079] The morphological difference analysis module is used to obtain the process reference sequences corresponding to multiple production processes of power equipment, construct the observation sequence of each normal sample and abnormal sample in each production process, and calculate multiple morphological difference features of each observation sequence based on the process reference sequence.

[0080] The difference extraction module is used to construct multiple process difference sequences for each observation sequence based on multiple morphological difference features, and to determine the start and end differences of each observation sequence in each production process based on the process difference sequences.

[0081] The difference propagation analysis module is used to integrate multiple initial and final differences, construct a difference state adjacency sequence between any two adjacent production processes based on the normal sample set, and calculate the difference propagation parameters between any two adjacent production processes based on multiple difference state adjacency sequences.

[0082] The anomaly propagation pattern recognition module is used to calculate multiple difference contribution factors for each anomaly sample in the defect sample set based on multiple difference propagation parameters, extract the difference propagation deviation features of each anomaly sample, identify multiple production anomaly propagation patterns in the defect sample set, extract multiple difference propagation fingerprint features, and use them for real-time production detection of power equipment.

[0083] The above are merely specific embodiments of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for intelligent analysis of power equipment production data, characterized in that, include: Collect production observation data of power equipment, classify the production observation data to generate multiple sets of normal samples and abnormal samples, and construct normal sample set and defect sample set of power equipment. Obtain the process reference sequences corresponding to the power equipment in multiple production processes, construct the observation sequences of each normal sample and abnormal sample in each production process, and calculate multiple morphological difference features of each observation sequence based on the process reference sequences. Multiple process difference sequences are constructed for each observation sequence based on multiple morphological difference characteristics. The starting and ending differences of each observation sequence in each production process are determined based on the process difference sequences. By integrating multiple initial and final differences, a difference state adjacency sequence between any two adjacent production processes is constructed based on the normal sample set. The difference propagation parameter between any two adjacent production processes is calculated based on the multiple difference state adjacency sequences. Based on multiple difference transmission parameters, calculate multiple difference contribution factors for each abnormal sample in the defect sample set. The difference contribution factors are used to represent the magnitude of the overall difference contribution of the process to the defect sample. Extract the difference transmission deviation features of each abnormal sample. Perform abnormal transmission pattern clustering on the abnormal transmission sequence constructed based on the difference contribution factors and difference transmission deviation features to identify multiple production abnormal transmission patterns in the defect sample set. An abnormal transition path is constructed for each production anomaly transmission mode. The abnormal transition path includes multiple transition points. Based on the end differences of multiple abnormal samples in each production process and the multiple difference transmission deviation characteristics of each abnormal sample, the transition intensity of each transition point is calculated. A transition intensity sequence of the abnormal transition path is constructed based on the transition intensity of the transition points. The transition intensity sequence is used to reflect the change characteristics of difference transmission of multiple abnormal samples in the production anomaly transmission mode throughout the entire production process. By combining the normal sample set, the abnormal intensity index of multiple abnormal samples in each production process in the production abnormality transmission pattern is calculated, and the abnormal intensity sequence of the production abnormality transmission pattern is generated. The transition intensity sequence and the abnormal intensity sequence are combined to generate the differential transmission fingerprint feature of the production abnormality transmission pattern. After acquiring real-time production monitoring data of power equipment, the real-time difference features of the real-time production monitoring data in each production process are extracted, and a real-time process difference sequence of the real-time production monitoring data is constructed. The real-time transmission intensity parameter of the real-time production monitoring data between any two adjacent production processes is calculated by combining multiple difference transmission parameters, and a real-time transmission intensity sequence of the real-time production monitoring data is constructed. Real-time production monitoring data is matched with multiple difference transmission fingerprint features based on real-time process difference sequences and real-time transmission intensity sequences to generate real-time production monitoring results for power equipment.

2. The intelligent analysis method for power equipment production data according to claim 1, characterized in that, Based on multiple morphological difference characteristics, multiple process difference sequences are constructed for each observation sequence. The starting and ending differences for each observation sequence at each production process are determined based on these process difference sequences, including: The morphological difference feature is the distance value between the process reference sequence and the observation sequence under the observation window. This includes performing a sliding window process based on timestamp alignment on the process reference sequence and the observation sequence, generating a reference sequence segment and an observation sequence segment under each observation window, and calculating the distance value between the reference sequence segment and the observation sequence segment to obtain the morphological difference feature of the observation sequence. Multiple morphological difference features of each production process are spliced ​​in time to construct the process difference sequence of each observation sequence in each production process. Based on the pre-set start state window and end state window, the sequence start segment and sequence end segment of each process difference sequence are extracted respectively. The start difference and end difference of each normal sample and abnormal sample are extracted from the sequence start segment and sequence end segment.

3. The intelligent analysis method for power equipment production data according to claim 2, characterized in that, By integrating multiple initial and final differences, a difference state adjacency sequence between any two adjacent production processes is constructed based on the normal sample set. The difference propagation parameters between any two adjacent production processes are calculated based on these multiple difference state adjacency sequences, including: Based on the production sequence of power equipment for multiple production processes, the starting and ending differences of multiple observation sequences in the normal sample set for each production process are combined to generate a difference propagation sample for each observation sequence. Multiple difference transmission samples between two adjacent production processes are spliced ​​and fused to construct a difference state adjacency sequence between two adjacent production processes. The difference state adjacency sequence includes an end state sequence and an start state sequence. The correlation coefficient between the end state sequence and the start state sequence is calculated to obtain the difference transmission parameters between any two adjacent production processes in the normal sample set.

4. The intelligent analysis method for power equipment production data according to claim 3, characterized in that, Based on multiple difference propagation parameters, multiple difference contribution factors are calculated for each abnormal sample in the defect sample set. Difference propagation deviation characteristics of each abnormal sample are extracted, and multiple production anomaly propagation patterns in the defect sample set are identified, including: Determine the initial and final differences for each outlier sample in each production process. Based on multiple difference propagation coefficients and final differences, calculate the unpropagated differences for each production process and record them as difference contribution factors. Based on the initial and final differences, the observed transfer parameters between any two adjacent production processes in the abnormal sample are calculated. Combined with the difference transfer parameters between the two production processes, multiple difference transfer deviation characteristics of each abnormal sample are calculated. By combining the difference contribution factor and the difference transmission deviation feature, an anomaly transmission sequence is constructed for each anomaly sample. Based on multiple anomaly transmission sequences, multiple groups of anomaly samples are clustered to generate multiple production anomaly transmission patterns in the defect sample set.

5. The intelligent analysis method for power equipment production data according to claim 1, characterized in that, Based on the real-time process difference sequence and the real-time transmission intensity sequence, real-time production monitoring data is matched with multiple difference transmission fingerprint features to generate real-time production monitoring results for power equipment, including: Calculate the abnormal intensity matching parameter between the real-time process difference sequence and the abnormal intensity sequence in the difference transmission fingerprint feature, calculate the abnormal transmission intensity matching parameter between the real-time transmission intensity sequence and the transition intensity sequence in the difference transmission fingerprint feature, fuse the abnormal intensity matching parameter and the abnormal transmission intensity matching parameter to obtain the abnormal pattern matching value between the real-time production monitoring data and multiple difference transmission fingerprint features, and generate the real-time production monitoring result of the power equipment based on the multiple abnormal pattern matching values.

6. A smart analysis system for power equipment production data, characterized in that, The system is used to implement the intelligent analysis method for power equipment production data as described in any one of claims 1-5, including: The production data acquisition module is used to collect production observation data of power equipment, classify the production observation data to generate multiple sets of normal samples and abnormal samples, and construct normal sample set and defect sample set of power equipment. The morphological difference analysis module is used to obtain the process reference sequences corresponding to multiple production processes of power equipment, construct the observation sequence of each normal sample and abnormal sample in each production process, and calculate multiple morphological difference features of each observation sequence based on the process reference sequence. The difference extraction module is used to construct multiple process difference sequences for each observation sequence based on multiple morphological difference features, and to determine the start and end differences of each observation sequence in each production process based on the process difference sequences. The difference propagation analysis module is used to integrate multiple initial and final differences, construct a difference state adjacency sequence between any two adjacent production processes based on the normal sample set, and calculate the difference propagation parameters between any two adjacent production processes based on multiple difference state adjacency sequences. The anomaly propagation pattern recognition module is used to calculate multiple difference contribution factors for each anomaly sample in the defect sample set based on multiple difference propagation parameters. The difference contribution factors are used to represent the magnitude of the overall difference contribution of the process to the defect sample. The module extracts the difference propagation deviation features of each anomaly sample, performs anomaly propagation pattern clustering on the anomaly propagation sequence constructed based on the difference contribution factors and difference propagation deviation features to identify multiple production anomaly propagation patterns in the defect sample set, and extracts multiple difference propagation fingerprint features based on the production anomaly propagation patterns for real-time production detection of power equipment.

Images