Power equipment safety early warning method and system based on multimodal data

Abstract

The present invention provides a power equipment safety early warning method and system based on multimodal data, which relates to the field of power equipment early warning, including: environmental attribute deviation vector, audio attribute deviation vector and operation attribute deviation vector, constructing spatial reference distribution coordinates, performing case retrieval in the power equipment safety management case library, constructing a recursive network tree for node weight distribution, and obtaining node weight distribution results; performing reverse growth aggregation on the recursive network tree according to the node weight distribution results, obtaining the power equipment abnormality index, and executing power equipment early warning when it is greater than or equal to the power equipment abnormality index threshold. The method solves the technical problem that due to the lack of a comprehensive analysis solution for multimodal data in the prior art, it is impossible to effectively use data of multiple attributes for comprehensive judgment, thereby making it difficult to achieve accurate early warning of power equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power equipment early warning, and in particular to an electric power equipment safety early warning method and system based on multimodal data. Background Art

[0002] During the operation of power equipment, equipment failure may lead to serious safety problems. With the complexity and intelligence of power systems, how to use effective methods to monitor and warn power equipment in real time to avoid potential equipment failures and safety hazards has become an important research direction. Traditional power equipment safety monitoring methods usually rely on single sensor data and cannot fully and accurately reflect the true status of the equipment. This makes it difficult to identify equipment anomalies in time before a failure occurs, making it impossible to effectively prevent safety accidents.

[0003] Existing technologies mainly rely on a single data source to monitor and warn power equipment. However, these single data sources often have limitations and cannot fully reflect the overall status of equipment operation. Especially when dealing with complex and changing environments, this single data source monitoring method is easily affected by noise interference and false alarms, resulting in insufficient accuracy and reliability of warnings. Summary of the invention

[0004] This application provides a power equipment safety early warning method and system based on multimodal data, aiming to solve the technical problem in the prior art that due to the lack of a comprehensive analysis solution for multimodal data, data with multiple attributes cannot be effectively used for comprehensive judgment, making it difficult to achieve accurate early warning of power equipment.

[0005] The first aspect disclosed in the present application provides a method for power equipment safety early warning based on multimodal data, including: calculating an environmental attribute vector threshold and an environmental attribute deviation vector of an environmental attribute monitoring vector, calculating an audio attribute vector threshold and an audio attribute deviation vector of an audio attribute monitoring vector, calculating an operation attribute vector threshold and an operation attribute deviation vector of an operation attribute monitoring vector; constructing a spatial reference distribution coordinate according to the environmental attribute deviation vector, the audio attribute deviation vector and the operation attribute deviation vector; performing case retrieval in a power equipment safety management case library according to the spatial reference distribution coordinates, and constructing a recursive network tree, wherein the power equipment safety management case library is used to store power equipment safety monitoring cases; performing node weight distribution according to the recursive network tree to obtain a node weight distribution result; performing reverse growth aggregation on the recursive network tree according to the node weight distribution result to obtain a power equipment abnormality index; and executing a power equipment early warning when the power equipment abnormality index is greater than or equal to the power equipment abnormality index threshold.

[0006] Another aspect disclosed in the present application provides an electric power equipment safety early warning system based on multimodal data, including: an attribute deviation vector calculation module, used to calculate an environmental attribute vector threshold and an environmental attribute deviation vector of an environmental attribute monitoring vector, calculate an audio attribute vector threshold and an audio attribute deviation vector of an audio attribute monitoring vector, and calculate an operation attribute vector threshold and an operation attribute deviation vector of an operation attribute monitoring vector; an encoding dimension reduction module, used to construct a spatial reference distribution coordinate according to the environmental attribute deviation vector, the audio attribute deviation vector and the operation attribute deviation vector; a recursive network tree construction module, used to perform case retrieval in an electric power equipment safety management case library according to the spatial reference distribution coordinates and construct a recursive network tree, wherein the electric power equipment safety management case library is used to store electric power equipment safety monitoring cases; a node weight distribution module, used to perform node weight distribution according to the recursive network tree to obtain a node weight distribution result; a reverse growth aggregation module, used to perform reverse growth aggregation on the recursive network tree according to the node weight distribution result to obtain an electric power equipment abnormality index; an electric power equipment early warning module, used to execute an electric power equipment early warning when the electric power equipment abnormality index is greater than or equal to the electric power equipment abnormality index threshold.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] By adopting the method of constructing spatial reference distribution coordinates through comprehensive analysis of deviation vectors of environmental attributes, audio attributes and operation attributes, the comprehensive monitoring and early warning of the power equipment status can be realized. The method first calculates the deviation vectors of various attributes, and then uses these vectors to construct a multi-dimensional spatial reference distribution coordinate as the basis for case retrieval and network tree construction. By retrieving similar cases in the power equipment safety management case library and constructing a recursive network tree, the method can conduct in-depth analysis of the operating status of the power equipment. The distribution of node weights and inverse growth aggregation further refine the identification of abnormal situations. Finally, by comparing the abnormal index of power equipment with the preset threshold, the technical solution of early warning of potential faults is realized. Through the fusion of multimodal data, the accuracy and reliability of the early warning system are improved. The constructed spatial reference distribution coordinates provide a more accurate reference framework for case retrieval, enabling the system to quickly locate and analyze potential equipment anomalies. The construction of the recursive network tree and the intelligent distribution of node weights enhance the system's adaptability and analysis capabilities to complex situations. The application of inverse growth aggregation rules optimizes the anomaly detection process and ensures the timeliness and effectiveness of early warning. Overall, this technical solution has significantly improved the efficiency and accuracy of power equipment safety warnings through innovative data fusion and intelligent analysis methods.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A possible flow chart of a method for safety early warning of electric power equipment based on multimodal data is provided for an embodiment of the present application.

[0011] Figure 2 A possible structural schematic diagram of a power equipment safety early warning system based on multimodal data is provided for an embodiment of the present application.

[0012] Explanation of the reference numerals: attribute deviation vector calculation module 1000, coding dimension reduction module 2000, recursive network tree construction module 3000, node weight distribution module 4000, reverse growth aggregation module 5000, power equipment early warning module 6000. DETAILED DESCRIPTION

[0013] Embodiment 1, as Figure 1 As shown, the embodiment of the present application provides a power equipment safety early warning method based on multimodal data, including:

[0014] Calculate the environmental attribute vector threshold and the environmental attribute deviation vector of the environmental attribute monitoring vector, calculate the audio attribute vector threshold and the audio attribute deviation vector of the audio attribute monitoring vector, and calculate the operating attribute vector threshold and the operating attribute deviation vector of the operating attribute monitoring vector.

[0015] Specifically, in the safety warning technology solution for power equipment, the environmental attribute vector, audio attribute vector, and operation attribute vector refer to a set of quantitative indicators that represent the environment, sound characteristics, and operation status of the power equipment. The threshold refers to the preset standard of these attributes under normal operating conditions. The deviation vector is the difference between the actual monitored attribute value and these preset thresholds. Calculating these deviation vectors is a key step in identifying whether the equipment deviates from the normal operating state.

[0016] First, standard thresholds for environmental attributes, audio attributes, and operational attributes are determined based on the attribute values ​​of the device in an ideal or normal state. Then, the actual values ​​of these attributes are monitored in real time and compared with the thresholds to calculate the deviation of each attribute. These deviations reflect the difference between the current state of the device and the ideal state, and are the basis for subsequent analysis and early warning.

[0017] The calculation of the deviation vector provides basic data for the subsequent construction of the spatial reference distribution coordinates. Through these deviation vectors, the abnormality of the equipment status can be quantified, which in turn provides input for case retrieval and the construction of the recursive network tree. These deviation vectors are the direct basis for judging whether the equipment needs early warning, and are the key link to achieve early detection of potential problems and prevention of failures.

[0018] A spatial reference distribution coordinate is constructed according to the environmental attribute deviation vector, the audio attribute deviation vector and the operation attribute deviation vector.

[0019] Specifically, in the technical solution for power equipment safety early warning, spatial reference distribution coordinates are a method to integrate multidimensional data into a unified reference frame. It combines environmental attribute deviation vectors, audio attribute deviation vectors, and operation attribute deviation vectors to form coordinate points in a multidimensional space, which represent a comprehensive snapshot of the equipment status.

[0020] The steps of constructing the spatial reference distribution coordinates involve mathematically processing and integrating the deviation vectors of three different attributes - environmental attributes, audio attributes, and operational attributes. This may include normalizing, weighting, or reducing the dimension of the deviation vectors through specific algorithms (such as principal component analysis PCA) to reduce the complexity of the data while retaining key information. Ultimately, these deviation vectors are converted into one or more coordinate points, and the positions of these coordinate points in multidimensional space can reflect the comprehensive status of the power equipment.

[0021] By positioning in multidimensional space, it is possible to make a more accurate comparison with historical cases, thereby improving the accuracy of early warning. In addition, the construction of spatial benchmark distribution coordinates is also a prerequisite for data fusion and intelligent analysis, laying a solid foundation for the efficient operation of the power equipment safety early warning system.

[0022] According to the spatial reference distribution coordinates, case retrieval is performed in a power equipment safety management case library to construct a recursive network tree, wherein the power equipment safety management case library is used to store power equipment safety monitoring cases.

[0023] Specifically, the spatial reference distribution coordinates are a multidimensional data point that integrates environmental, audio, and operational attribute deviations, and is used to characterize the real-time status of power equipment. Case retrieval refers to finding cases similar to the current equipment status in an existing case library for comparison and analysis. A recursive network tree is a data structure used to organize and represent the relationship between cases. It is constructed in a recursive manner and can efficiently store and retrieve cases. The power equipment safety management case library is a database that stores historical safety monitoring cases, which include monitoring data and results of equipment in different states.

[0024] This step involves using the spatial reference distribution coordinates as the query basis to retrieve cases in the power equipment safety management case library. The purpose of the retrieval is to find historical cases that are similar to the current equipment status, which can provide references for analyzing the current equipment status. The retrieved cases will be used to build a recursive network tree, which will be organized according to the similarity or other correlations between cases. The construction of the recursive network tree helps the system better understand and learn the change pattern of the equipment status, thereby improving the accuracy and efficiency of early warning.

[0025] Node weight distribution is performed according to the recursive network tree to obtain a node weight distribution result.

[0026] Specifically, the node weight distribution result is a weight parameter that characterizes the importance of each node in the recursive network tree. By reasonably assigning weights to the nodes, the system can identify which cases or modes are more critical under the current device status, thereby more accurately predicting and assessing potential risks.

[0027] The recursive network tree is subjected to reverse growth aggregation according to the node weight distribution result to obtain an abnormal index of power equipment.

[0028] Specifically, reverse growth aggregation is an analysis technique that starts from the bottom node of the recursive network tree and aggregates information layer by layer until the root node of the tree. The power equipment abnormality index is a quantitative indicator used to measure the degree of abnormality of the power equipment status. Reverse growth aggregation comprehensively considers the weight of each node and finally calculates the power equipment abnormality index. Details are as follows:

[0029] First, determine the number of positive and negative examples for each child node, and perform weighted summation based on the node weight distribution results to calculate the child node score. The score reflects the importance and abnormality of the child node to the parent node. Then, adjust the number of positive or negative examples of the parent node based on the positive example score ratio threshold. This process continues until the root node, and finally aggregates to obtain the power equipment abnormality index.

[0030] The abnormal index of power equipment provides a quantitative basis for early warning decisions. When the abnormal index exceeds the preset threshold, the early warning mechanism is triggered and timely measures are taken. This method improves the response speed and accuracy of the early warning system, helps prevent failures and accidents, and ensures the safe and stable operation of the power system.

[0031] When the power equipment abnormality index is greater than or equal to the power equipment abnormality index threshold, a power equipment early warning is executed.

[0032] Specifically, when the calculation result of the power equipment abnormality index exceeds or reaches the preset threshold, it indicates that the equipment may have a risk of failure or abnormality. At this time, the system will execute the early warning procedure, which may include issuing an alarm, notifying relevant personnel, recording event logs, automatically adjusting equipment operating parameters, or taking other emergency measures to ensure the safe operation of the power system.

[0033] Further, constructing a spatial reference distribution coordinate according to the environmental attribute deviation vector, the audio attribute deviation vector and the operation attribute deviation vector includes:

[0034] Perform dimension reduction encoding on the environmental attribute deviation vector to obtain the environmental attribute deviation encoding.

[0035] Perform dimension reduction encoding on the audio attribute deviation vector to obtain the audio attribute deviation code.

[0036] Perform dimension reduction encoding on the operation attribute deviation vector to obtain the operation attribute deviation encoding.

[0037] The spatial reference distribution coordinates are constructed according to the environmental attribute deviation code, the audio attribute deviation code and the operation attribute deviation code.

[0038] First, the environmental attribute deviation vector, audio attribute deviation vector, and operation attribute deviation vector are processed by dimensionality reduction coding to generate the corresponding environmental attribute deviation code, audio attribute deviation code, and operation attribute deviation code. Dimensionality reduction coding is a data compression technology that simplifies the data structure by reducing the data dimension while retaining important feature information as much as possible. Subsequently, the three codes are combined to construct the spatial reference distribution coordinates.

[0039] Dimensionality reduction coding effectively reduces the complexity of data and improves the efficiency of data processing, while retaining the characteristic information that is critical to early warning analysis. This approach allows the system to perform more sophisticated case matching and anomaly detection in a higher-dimensional data space, thereby improving the overall performance and reliability of the early warning system.

[0040] Further, the environmental attribute deviation vector is subjected to dimensionality reduction encoding to obtain the environmental attribute deviation encoding, including:

[0041] Configure environmental attributes and the minimum scale for environmental attribute monitoring.

[0042] A high-dimensional coordinate space is constructed with the environmental attributes.

[0043] A space partition is constructed by monitoring the minimum scale of the environmental attribute, wherein each edge of the space partition corresponds to the environmental attribute one by one.

[0044] Based on the space partitioning body, the high-dimensional coordinate space is partitioned from the origin to obtain a dimension-reduced coding grid space.

[0045] Set the coding symbol library and coding distribution rules.

[0046] Based on the coding symbol library and the coding distribution rule, the environmental attribute deviation vector is unsupervisedly trained in the dimension reduction coding grid space to obtain the environmental attribute deviation coding.

[0047] First, it is necessary to configure the environmental attributes and the minimum scale of their monitoring. For example, if the temperature accuracy is 0.01°C, the minimum scale is 0.01°C. Then, use these environmental attributes to construct a high-dimensional coordinate space, which can comprehensively represent the various dimensions of the environmental attributes. Then, based on the minimum scale of monitoring, as the side length, a high-dimensional space partition is constructed. Each side of this partition corresponds to an environmental attribute, thereby dividing the high-dimensional space into multiple subspaces. Further, based on this space partition, the high-dimensional coordinate space is partitioned from the origin, and the high-dimensional coordinate space is stored as a reduced-dimensional coding grid space after the partition. Setting the coding symbol library refers to a preset set of non-repeated coding characters, and the coding distribution rule refers to distributing the coding character set to the reduced-dimensional coding grid space, distributing one character in each subspace, and not repeating, to ensure that the last character can represent the position in the high-dimensional coordinate space. By setting the coding symbol library and the coding distribution rule, the environmental attribute deviation vector is unsupervisedly trained in the grid space, and finally the environmental attribute deviation coding is obtained.

[0048] In detail, first, based on the above rules, after the dimensionality reduction coding grid space is constructed, a preset number of historical data of environmental attribute deviation vectors are collected, and the historical data of environmental attribute deviation vectors are trained based on the set coding symbol library and coding distribution rules. When at least 85 out of 100 consecutive training times are accurate, the unsupervised training is considered to have converged, and a dimensionality reduction encoder is generated. The environmental attribute deviation vector is processed to obtain the environmental attribute deviation coding.

[0049] Further, according to the spatial reference distribution coordinates, case retrieval is performed in the power equipment safety management case library to construct a recursive network tree, which includes:

[0050] When the preset update cycle is met, the power equipment safety monitoring affairs are collected online according to the power equipment model, wherein the power equipment safety monitoring affairs include environmental attribute deviation vector record data, audio attribute deviation vector record data, operation attribute deviation vector record data and fault type code record data.

[0051] The environmental attribute deviation vector record data, the audio attribute deviation vector record data, and the operation attribute deviation vector record data are respectively dimensionality reduced and encoded to obtain environmental attribute deviation encoded record data, audio attribute deviation encoded record data, and operation attribute deviation encoded record data, and to construct spatial identification distribution coordinates.

[0052] When the fault type coding record data is 0, a positive case of power equipment safety management is constructed according to the spatial identification distribution coordinates and added to the power equipment safety management case library.

[0053] When the fault type coding record data is other, a negative case for safety management of electric equipment is constructed according to the spatial identification distribution coordinates and added to the negative case for safety management of electric equipment.

[0054] Among them, when the storage time of the negative case of power equipment safety management or the positive case of power equipment safety management meets the preset deletion time, the negative case of power equipment safety management or the positive case of power equipment safety management is deleted from the power equipment safety management case library.

[0055] Specifically, the power equipment safety monitoring transaction refers to a series of monitoring data related to the safety of the power equipment, including the deviation vector of the environment, audio, and operation attributes and the fault type code. The fault type code is used to identify whether the power equipment has a fault and the fault type code, where 0 usually means no fault. When the preset update cycle is met, the system will collect relevant safety monitoring transaction data through the network according to the model of the power equipment. These data include environmental attribute deviation vectors, audio attribute deviation vectors, operation attribute deviation vectors, and fault type codes. Then, dimensionality reduction coding is performed on each type of deviation vector record data collected to generate corresponding coded record data, and these data are used to construct spatial identification distribution coordinates. According to the value of the fault type code, the monitoring transaction is classified into a positive case (no fault, coded as 0) or a negative case (there is a fault, coded as other values), and added to the safety management case library respectively. In addition, the cases in the case library will be managed according to the preset deletion time, and the timed cases will be removed from the library. Preferably, the dimensionality reduction coding here is also processed in the same way as mentioned above, and no further details will be given here.

[0056] By distinguishing between positive and negative cases, the system can learn the device behavior patterns under normal and abnormal conditions, thereby improving the recognition ability of the early warning system. The construction of the recursive network tree provides a hierarchical storage and retrieval method for cases, making the relationship between cases clearer and helping to quickly locate and respond to potential safety issues. In addition, by regularly updating and maintaining the case library, it ensures that the case library data is up to date and reflects the actual operation of the equipment, thereby improving the reliability and practicality of the early warning system.

[0057] Further, according to the spatial reference distribution coordinates, case retrieval is performed in the power equipment safety management case library to construct a recursive network tree, including:

[0058] Configure the first-order distance threshold, the second-order distance threshold, and up to the K-order distance threshold, where K is an integer, K≥20.

[0059] According to the spatial reference distribution coordinates, a first-order case set that meets the first-order distance threshold is retrieved from the power equipment safety management case library.

[0060] The first-order case set is traversed, and a second-order case set that meets the second-order distance threshold is searched in the power equipment safety management case library.

[0061] Until the K-1 order case set is traversed, a K order case set satisfying the K order distance threshold is retrieved from the power equipment safety management case library.

[0062] The spatial reference distribution coordinates are set as the root node, the first-order case set is set as the first-level leaf node, the second-order case set is set as the second-level leaf node, until the K-order case set is set as the K-level leaf node, to construct the recursive network tree.

[0063] Specifically, the preset value of the distance threshold is used to measure the degree of proximity between the case and the reference coordinates in the multidimensional space, including the first-order distance threshold, the second-order distance threshold, and up to the K-order distance threshold. The first-order case set / second-order case set up to the K-order case set, the case set retrieved according to the distance threshold, corresponds to the distance thresholds of different orders, indicating the degree of proximity between the case and the reference coordinates. The manager pre-sets a series of distance thresholds, from the first-order distance threshold to the K-order distance threshold, where K is an integer greater than or equal to 20, and the specific distance threshold needs to be set in combination with the actual power equipment application scenario. These distance thresholds are used to determine the similarity or proximity between the case and the spatial reference distribution coordinates. When it is less than or equal to the distance threshold, it is deemed to meet the distance threshold. Next, the first-order case set that meets the distance threshold with the spatial reference distribution coordinates is retrieved from the case library. Subsequently, the first-order case set is traversed, and the second-order case set is retrieved using the second-order distance threshold until the K-1 order case set is traversed, and the K-order case set that meets the K-order distance threshold is retrieved in the power equipment safety management case library; the spatial reference distribution coordinates are set as the root node, the first-order case set is set as the first-level leaf node, and the second-order case set is set as the second-level leaf node, until the K-order case set is set as the K-level leaf node, and the recursive network tree is constructed. The recursive network tree can hierarchically represent the relationship between the case and the reference coordinates, making case retrieval more efficient and targeted. Secondly, by setting multi-order distance thresholds, the system can flexibly adjust the precision of the retrieval to adapt to different monitoring needs and early warning scenarios.

[0064] Further, performing node weight distribution according to the recursive network tree to obtain a node weight distribution result includes:

[0065] According to the recursive network tree, the first child node, the second child node, and finally the Mth child node of the i-th node are obtained.

[0066] A first node distance between the i-th node and the first child node, a second node distance between the i-th node and the second child node, and so on, are obtained, until an M-th node distance between the i-th node and the M-th child node.

[0067] The first node distance, the second node distance, and finally the Mth node distance are added to obtain a distance sum result.

[0068] The first node distance, the second node distance, and the Mth node distance are traversed, and compared with the sum of the distances to obtain the first child node weight, the second child node weight, and the Mth child node weight.

[0069] Specifically, the i-th node refers to a specific node being analyzed in the recursive network tree. The child node refers to the next-level node directly connected to the i-th node, including the first child node to the M-th child node. Determine all the child nodes of any i-th node in the recursive network tree, which are from the first child node to the M-th child node. Next, calculate the distance between the i-th node and each child node, and these distances are respectively called the first node distance to the M-th node distance. Then, add up these distance values ​​to obtain a distance sum result. After that, calculate the weight of each child node relative to other child nodes by comparing the distance of each child node with the sum result. This process involves traversing the distances of all child nodes and determining the weights based on their ratio to the sum. The distribution of weights helps to perform more accurate case matching and anomaly detection in the recursive network tree, because nodes with higher weights may represent more typical cases or patterns.

[0070] Further, the recursive network tree is subjected to reverse growth aggregation according to the node weight distribution result to obtain an abnormal index of power equipment, including:

[0071] Construct reverse growth aggregation rules:

[0072] Obtain the first child node, the second child node, and so on until the Qth child node of the jth node, wherein the first child node has a first number of positive examples and a first number of negative examples, the second child node has a second number of positive examples and a second number of negative examples, and so on until the Qth child node has a Qth number of positive examples and a Qth number of negative examples.

[0073] According to the node weight distribution result, weighted sum is performed on the first number of positive examples and the first number of negative examples, the second number of positive examples and the second number of negative examples, and the Qth number of positive examples and the Qth number of negative examples to obtain positive example scores and negative example scores, and the total score is calculated.

[0074] When the ratio of the positive example score to the total score is less than or equal to the positive example score ratio threshold, the number of negative examples of the j-th node is increased by one.

[0075] When the ratio of the positive example score to the total score is greater than the positive example score ratio threshold, the number of positive examples of the j-th node is increased by one.

[0076] According to the reverse growth aggregation rule, the recursive network tree is aggregated from the bottom layer to the root node to obtain the root node case type.

[0077] When the root node case type is a positive case, the power equipment abnormality index is equal to 0.

[0078] When the root node case type is a negative example, the power equipment abnormality index is equal to 1, and at the same time, statistics are counted for the trigger frequency of the recursive network tree, and the preset number of fault types are associated with the power equipment abnormality index to build push data.

[0079] Specifically, reverse growth aggregation refers to a bottom-up data processing method that starts from the bottom of the network tree and aggregates information layer by layer until the root node. Positive examples usually refer to cases without faults, while negative examples refer to cases with faults. The positive score ratio threshold refers to a preset ratio used to determine the case type of a node, i.e., faulty or non-faulty.

[0080] Constructing the reverse growth aggregation rule: Analyze all the child nodes of the jth node (from the first child node to the Qth child node), and each child node has its corresponding number of positive and negative examples. Then, according to the node weight distribution result, the number of positive and negative examples of all child nodes is weighted and summed to obtain the positive example score and the negative example score, and the total score is calculated. Next, according to the comparison result of the ratio of the positive example score to the total score and the positive example score ratio threshold, the number of positive or negative examples of the jth node is increased accordingly. According to the reverse growth aggregation rule, aggregation is performed from the bottom layer to the root node until the root node case type is obtained. When the root node case type is a positive example, the power equipment abnormality index is equal to 0; when the root node case type is a negative example, the power equipment abnormality index is equal to 1. At the same time, the trigger frequency of the recursive network tree is counted. The preset number of fault types and the power equipment abnormality index are associated with the push data.

[0081] In summary, the electric power equipment safety early warning method and system based on multimodal data provided by the embodiments of the present application have the following technical effects:

[0082] 1. By comprehensively analyzing the deviation vectors of environmental attributes, audio attributes and operating attributes, the spatial reference distribution coordinates are constructed to achieve comprehensive monitoring and early warning of the power equipment status. The method first calculates the deviation vectors of various attributes, and then uses these vectors to construct a multidimensional spatial reference distribution coordinate as the basis for case retrieval and network tree construction. By retrieving similar cases in the power equipment safety management case library and constructing a recursive network tree, the method can conduct an in-depth analysis of the operating status of the power equipment. The distribution of node weights and inverse growth aggregation further refine the identification of abnormal situations. Finally, by comparing the abnormal index of the power equipment with the preset threshold, the technical solution for early warning of potential faults is realized. Through the fusion of multimodal data, the accuracy and reliability of the early warning system are improved. The constructed spatial reference distribution coordinates provide a more accurate reference framework for case retrieval, enabling the system to quickly locate and analyze potential equipment anomalies. The construction of the recursive network tree and the intelligent distribution of node weights enhance the system's adaptability and analysis capabilities to complex situations. The application of inverse growth aggregation rules optimizes the anomaly detection process and ensures the timeliness and effectiveness of early warning. Overall, this technical solution has significantly improved the efficiency and accuracy of power equipment safety warnings through innovative data fusion and intelligent analysis methods.

[0083] Embodiment 2 is based on the same inventive concept as the power equipment safety early warning method based on multimodal data in the above embodiment. Figure 2 As shown, the embodiment of the present application provides a power equipment safety early warning system based on multimodal data, including:

[0084] The attribute deviation vector calculation module 1000 is used to calculate the environmental attribute deviation vector of the environmental attribute vector threshold and the environmental attribute monitoring vector, calculate the audio attribute deviation vector of the audio attribute vector threshold and the audio attribute monitoring vector, and calculate the operation attribute deviation vector of the operation attribute vector threshold and the operation attribute monitoring vector.

[0085] The coding dimension reduction module 2000 is used to construct a spatial reference distribution coordinate according to the environmental attribute deviation vector, the audio attribute deviation vector and the operation attribute deviation vector.

[0086] The recursive network tree construction module 3000 is used to perform case retrieval in the power equipment safety management case library according to the spatial reference distribution coordinates to construct a recursive network tree, wherein the power equipment safety management case library is used to store power equipment safety monitoring cases.

[0087] The node weight distribution module 4000 is used to perform node weight distribution according to the recursive network tree to obtain a node weight distribution result.

[0088] The reverse growth aggregation module 5000 is used to perform reverse growth aggregation on the recursive network tree according to the node weight distribution result to obtain the power equipment abnormality index.

[0089] The electric power equipment early warning module 6000 is used to execute an electric power equipment early warning when the electric power equipment abnormality index is greater than or equal to the electric power equipment abnormality index threshold.

[0090] Furthermore, the coding and dimensionality reduction module 2000 executes the steps of: constructing spatial reference distribution coordinates according to the environmental attribute deviation vector, the audio attribute deviation vector and the operating attribute deviation vector, including: performing dimensionality reduction encoding on the environmental attribute deviation vector to obtain an environmental attribute deviation code; performing dimensionality reduction encoding on the audio attribute deviation vector to obtain an audio attribute deviation code; performing dimensionality reduction encoding on the operating attribute deviation vector to obtain an operating attribute deviation code; constructing the spatial reference distribution coordinates according to the environmental attribute deviation code, the audio attribute deviation code and the operating attribute deviation code.

[0091] Furthermore, the coding dimension reduction module 2000 executes the steps of: performing dimension reduction coding on the environmental attribute deviation vector to obtain the environmental attribute deviation coding, including: configuring environmental attributes and the minimum scale for monitoring environmental attributes; constructing a high-dimensional coordinate space with the environmental attributes; constructing a space partition body with the minimum scale for monitoring environmental attributes, wherein each edge of the space partition body corresponds one-to-one to the environmental attribute; based on the space partition body, splitting the high-dimensional coordinate space from the origin to obtain a dimension reduction coding grid space; setting a coding symbol library and a coding distribution rule; based on the coding symbol library and the coding distribution rule, performing unsupervised training on the environmental attribute deviation vector in the dimension reduction coding grid space to obtain the environmental attribute deviation coding.

[0092] Furthermore, the recursive network tree construction module 3000 executes the steps of: performing case retrieval in the power equipment safety management case library according to the spatial reference distribution coordinates to construct a recursive network tree, which includes: when a preset update cycle is met, according to the power equipment model, collecting power equipment safety monitoring affairs online, wherein the power equipment safety monitoring affairs include environmental attribute deviation vector record data, audio attribute deviation vector record data, operation attribute deviation vector record data and fault type coding record data; performing dimensionality reduction coding on the environmental attribute deviation vector record data, the audio attribute deviation vector record data and the operation attribute deviation vector record data to obtain environmental attribute deviation coding record data, audio attribute deviation vector record data and operation attribute deviation vector record data respectively. The fault type coding record data and the operation attribute deviation coding record data are used to construct the spatial identification distribution coordinates; when the fault type coding record data is 0, a positive case of power equipment safety management is constructed according to the spatial identification distribution coordinates, and added to the power equipment safety management case library; when the fault type coding record data is other, a negative case of power equipment safety management is constructed according to the spatial identification distribution coordinates, and added to the power equipment safety management case library; wherein, when the storage time of the negative case of power equipment safety management or the positive case of power equipment safety management meets the preset deletion time, the negative case of power equipment safety management or the positive case of power equipment safety management is deleted from the power equipment safety management case library.

[0093] Further, the recursive network tree construction module 3000 executes the steps of: performing case retrieval in the electric power equipment safety management case library according to the spatial reference distribution coordinates, and constructing a recursive network tree, including: configuring a first-order distance threshold, a second-order distance threshold, and up to a K-order distance threshold, wherein K is an integer, and K≥20; searching the electric power equipment safety management case library for a first-order case set that meets the first-order distance threshold according to the spatial reference distribution coordinates; traversing the first-order case set, and searching the electric power equipment safety management case library for a second-order case set that meets the second-order distance threshold; until the K-1-order case set is traversed, and the electric power equipment safety management case library is searched for a K-order case set that meets the K-order distance threshold; setting the spatial reference distribution coordinates as the root node, the first-order case set as the first-level leaf node, and the second-order case set as the second-level leaf node, until the K-order case set is set as the K-level leaf node, and constructing the recursive network tree.

[0094] Further, the node weight distribution module 4000 executes the steps of: performing node weight distribution according to the recursive network tree to obtain a node weight distribution result, including: obtaining the first child node, the second child node, and so on of the i-th node according to the recursive network tree; obtaining the first node distance between the i-th node and the first child node, the second node distance between the i-th node and the second child node, and so on ...

[0095] Further, the reverse growth aggregation module 5000 executes the steps of: performing reverse growth aggregation on the recursive network tree according to the node weight distribution result to obtain the power equipment abnormality index, including: constructing a reverse growth aggregation rule: obtaining the first child node, the second child node until the Qth child node of the jth node, wherein the first child node has a first number of positive examples and a first number of negative examples, the second child node has a second number of positive examples and a second number of negative examples, until the Qth child node has a Qth number of positive examples and a Qth number of negative examples; according to the node weight distribution result, performing weighted summation on the first number of positive examples and the first number of negative examples, the second number of positive examples and the second number of negative examples, and the Qth number of positive examples and the Qth number of negative examples, The positive example score and the negative example score are obtained, and the total score is calculated; when the ratio of the positive example score to the total score is less than or equal to the positive example score ratio threshold, the number of negative examples of the j-th node is increased by one; when the ratio of the positive example score to the total score is greater than the positive example score ratio threshold, the number of positive examples of the j-th node is increased by one; according to the reverse growth aggregation rule, the recursive network tree is aggregated from the bottom layer to the root node to obtain the root node case type; when the root node case type is a positive example, the power equipment abnormality index is equal to 0; when the root node case type is a negative example, the power equipment abnormality index is equal to 1, and at the same time, the trigger frequency of the recursive network tree is counted. The preset number of fault types and the power equipment abnormality index are associated with push data.

[0096] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0097] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A power equipment safety early warning method based on multimodal data, characterized in that: include: Calculate the environmental attribute vector threshold and the environmental attribute deviation vector of the environmental attribute monitoring vector, calculate the audio attribute vector threshold and the audio attribute deviation vector of the audio attribute monitoring vector, and calculate the operating attribute vector threshold and the operating attribute deviation vector of the operating attribute monitoring vector; Constructing a spatial reference distribution coordinate according to the environmental attribute deviation vector, the audio attribute deviation vector and the operation attribute deviation vector; According to the spatial reference distribution coordinates, case retrieval is performed in a power equipment safety management case library to construct a recursive network tree, wherein the power equipment safety management case library is used to store power equipment safety monitoring cases; Performing node weight distribution according to the recursive network tree to obtain a node weight distribution result; According to the node weight distribution result, the recursive network tree is subjected to reverse growth aggregation to obtain an abnormal index of power equipment; When the power equipment abnormality index is greater than or equal to the power equipment abnormality index threshold, executing power equipment early warning; According to the spatial reference distribution coordinates, case retrieval is performed in the power equipment safety management case library to construct a recursive network tree, including: Configure the first-order distance threshold, the second-order distance threshold, and the K-order distance threshold, where K is an integer, K ≥ 20; According to the spatial reference distribution coordinates, searching the electric power equipment safety management case library for a first-order case set that meets the first-order distance threshold; Traversing the first-order case set, and searching the electric power equipment safety management case library for a second-order case set that meets the second-order distance threshold; Until the K-1 order case set is traversed, a K order case set satisfying the K order distance threshold is searched in the power equipment safety management case library; The spatial reference distribution coordinate is set as a root node, the first-order case set is set as a first-level leaf node, the second-order case set is set as a second-level leaf node, until the K-order case set is set as a K-level leaf node, to construct the recursive network tree; Performing node weight distribution according to the recursive network tree to obtain a node weight distribution result includes: According to the recursive network tree, obtain the first child node, the second child node, and the Mth child node of the i-th node; Obtaining a first node distance between the i-th node and the first child node, a second node distance between the i-th node and the second child node, and so on until an M-th node distance between the i-th node and the M-th child node; Add the first node distance, the second node distance, and up to the Mth node distance to obtain a distance sum result; Traversing the first node distance, the second node distance until the M-th node distance, and comparing with the sum of the distances to obtain the first child node weight, the second child node weight until the M-th child node weight; According to the node weight distribution result, the recursive network tree is subjected to reverse growth aggregation to obtain an abnormal index of power equipment, including: Construct reverse growth aggregation rules: Obtain a first child node, a second child node, and finally a Qth child node of the jth node, wherein the first child node has a first number of positive examples and a first number of negative examples, the second child node has a second number of positive examples and a second number of negative examples, and finally the Qth child node has a Qth number of positive examples and a Qth number of negative examples; According to the node weight distribution result, weighted sum the first number of positive examples and the first number of negative examples, the second number of positive examples and the second number of negative examples, and the Qth number of positive examples and the Qth number of negative examples are performed to obtain positive example scores and negative example scores, and calculate the total score; When the ratio of the positive example score to the total score is less than or equal to the positive example score ratio threshold, the number of negative examples of the j-th node is increased by one; When the ratio of the positive example score to the total score is greater than the positive example score ratio threshold, the number of positive examples of the j-th node is increased by one; According to the reverse growth aggregation rule, the recursive network tree is aggregated from the bottom to the root node to obtain the root node case type; When the root node case type is a positive case, the power equipment abnormality index is equal to 0; When the root node case type is a negative example, the power equipment abnormality index is equal to 1, and at the same time, statistics are counted for the trigger frequency of the recursive network tree, and the preset number of fault types are associated with the power equipment abnormality index to build push data.

2. The method according to claim 1, characterized in that Constructing a spatial reference distribution coordinate according to the environmental attribute deviation vector, the audio attribute deviation vector, and the operation attribute deviation vector, including: Performing dimension reduction encoding on the environmental attribute deviation vector to obtain an environmental attribute deviation encoding; Performing dimension reduction encoding on the audio attribute deviation vector to obtain an audio attribute deviation code; Performing dimension reduction encoding on the operation attribute deviation vector to obtain an operation attribute deviation code; The spatial reference distribution coordinates are constructed according to the environmental attribute deviation code, the audio attribute deviation code and the operation attribute deviation code.

3. The method according to claim 2, characterized in that Performing dimension reduction encoding on the environmental attribute deviation vector to obtain the environmental attribute deviation encoding includes: Configure environmental attributes and the minimum scale for environmental attribute monitoring; Constructing a high-dimensional coordinate space with the environmental attributes; Monitoring the minimum scale of the environmental attribute to construct a space partition, wherein each edge of the space partition corresponds to the environmental attribute one by one; Based on the space partitioning body, the high-dimensional coordinate space is partitioned from the origin to obtain a dimension-reduced coding grid space; Set up the coding symbol library and coding distribution rules; Based on the coding symbol library and the coding distribution rule, the environmental attribute deviation vector is unsupervisedly trained in the dimension reduction coding grid space to obtain the environmental attribute deviation coding.

4. The method according to claim 1, characterized in that According to the spatial reference distribution coordinates, case retrieval is performed in the power equipment safety management case library to construct a recursive network tree, which includes: When the preset update cycle is met, the power equipment safety monitoring affairs are collected online according to the power equipment model, wherein the power equipment safety monitoring affairs include environmental attribute deviation vector record data, audio attribute deviation vector record data, operation attribute deviation vector record data and fault type code record data; Performing dimensionality reduction encoding on the environmental attribute deviation vector record data, the audio attribute deviation vector record data, and the operation attribute deviation vector record data respectively, obtaining environmental attribute deviation encoding record data, audio attribute deviation encoding record data, and operation attribute deviation encoding record data, and constructing spatial identification distribution coordinates; When the fault type code record data is 0, construct a positive case of power equipment safety management according to the spatial identification distribution coordinates, and add it to the power equipment safety management case library; When the fault type code record data is other, construct a negative case of power equipment safety management according to the spatial identification distribution coordinates, and add it to the power equipment safety management case library; Among them, when the storage time of the negative case of power equipment safety management or the positive case of power equipment safety management meets the preset deletion time, the negative case of power equipment safety management or the positive case of power equipment safety management is deleted from the power equipment safety management case library.

5. The power equipment safety early warning system based on multimodal data is characterized by: A method for implementing a safety early warning method for electric power equipment based on multimodal data according to any one of claims 1 to 4, the system comprising: An attribute deviation vector calculation module is used to calculate an environment attribute vector threshold and an environment attribute deviation vector of an environment attribute monitoring vector, calculate an audio attribute vector threshold and an audio attribute deviation vector of an audio attribute monitoring vector, and calculate an operation attribute vector threshold and an operation attribute deviation vector of an operation attribute monitoring vector; A coding dimension reduction module, used for constructing a spatial reference distribution coordinate according to the environmental attribute deviation vector, the audio attribute deviation vector and the operation attribute deviation vector; A recursive network tree construction module, used to perform case retrieval in a power equipment safety management case library according to the spatial reference distribution coordinates, and to construct a recursive network tree, wherein the power equipment safety management case library is used to store power equipment safety monitoring cases; A node weight distribution module, used to perform node weight distribution according to the recursive network tree to obtain a node weight distribution result; A reverse growth aggregation module, used for performing reverse growth aggregation on the recursive network tree according to the node weight distribution result to obtain an abnormal index of power equipment; The electric power equipment early warning module is used to execute the electric power equipment early warning when the electric power equipment abnormality index is greater than or equal to the electric power equipment abnormality index threshold.

Images