A power material detection task generation method and device

Abstract

The application discloses a kind of electric power material detection task generation method and device, it is related to electric power material detection technical field, the specific steps of the method include: based on the analysis of the information of electric power material to be detected, give multiple target features, target feature is processed by electric power material clustering classification model, and the detection classification of electric power material is given;Combined with the detection classification of electric power material and the detection demand model constructed in advance, obtain candidate detection item set;Based on the detection method library established in advance, with multi-objective optimization function, and through the iteration optimization of genetic algorithm, give the optimal detection item set;Combined with real-time detection resource and time cost, and using dynamic adaptation degree field matching algorithm, for each detection item in the optimal detection item set Matching corresponding detection method, generating detection task list.The application realizes the efficient use of detection resource, and considers the detection accuracy, detection reliability and cost benefit of electric power material.

Description

Technical Field

[0001] This invention belongs to the field of power material testing technology, specifically relating to a method and apparatus for generating power material testing tasks. Background Technology

[0002] As a crucial link in ensuring the health status of power equipment, power material inspection faces challenges such as heavy inspection tasks, uneven resource allocation, and difficulty in balancing inspection efficiency and accuracy.

[0003] Traditional testing methods rely on manual experience to formulate testing plans, resulting in high subjectivity, insufficient coverage, and resource waste. Currently, for the testing of power materials, methods based on static rules have been developed for tracking and tracing the testing of power materials. For example, patent CN114092047A discloses a traceable system and method for power material testing information. This system is based on blockchain technology and includes a data layer, network layer, consensus layer, contract layer, and application layer. To trace power materials, the traceable system first obtains the power material testing information, verifies and encrypts it to generate a genesis block, then obtains the power material lifecycle information, verifies and encrypts it to generate other blocks. During the block generation process, the finality mechanism in the contract layer prevents malicious tampering of information. Finally, based on the genesis block and the other blocks, the entire lifecycle information of the power material can be traced. The design of system identification IDs and human identification IDs ensures that the physical power materials offline match the system identification IDs online, facilitating direct access to online data by offline staff.

[0004] Furthermore, power material inspection involves multiple types of equipment, numerous inspection items, and complex scenarios, placing higher demands on the dynamic management of inspection cycles, indicator thresholds, and related rules. Relying on manual experience or static rules has the following drawbacks: First, insufficient utilization of historical inspection data and a lack of in-depth analysis and standardized processing of various report types lead to data fragmentation and low efficiency in extracting key information. Second, the construction of inspection demand models is crude, failing to fully consider material status characteristics, historical anomaly data, and rule correlations, easily resulting in over-inspection or under-inspection. Third, the selection of inspection items lacks a multi-objective optimization mechanism, making it difficult to balance comprehensiveness, resource adaptability, and time efficiency, resulting in insufficient rationality of inspection plans. Finally, the matching of inspection methods and resources relies on fixed rules, failing to dynamically adapt to real-time resource status and lacking flexible adjustment strategies when resource conflicts occur. In addition, traditional methods cannot automatically generate inspection task lists, requiring manual information integration, which is inefficient and prone to errors.

[0005] Therefore, there is an urgent need for an intelligent and systematic task generation method to achieve optimized allocation of detection resources and precise matching of detection needs. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a task generation method and device for power material inspection. By constructing a semantic tensor extraction model for power materials, it achieves in-depth analysis and standardized storage of historical inspection reports; by combining an inspection demand model, clustering classification algorithm and multi-objective optimization function, it accurately predicts inspection demand and selects the optimal inspection items; and by using a dynamic adaptation field algorithm to match inspection methods and resources, it generates an executable inspection task list.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, a task generation method for power material detection, the specific steps of which are as follows:

[0008] Based on the analysis of information on power materials to be detected, multiple target features are given. The target features are then processed through a power material clustering classification model to provide a detection classification of the power materials.

[0009] By combining the inspection classification of power materials with the pre-built inspection demand model, a set of candidate inspection items is obtained;

[0010] Based on a pre-established detection method library, an optimal set of detection items is given by using a multi-objective optimization function and iterative optimization through a genetic algorithm.

[0011] By combining real-time detection resources and time costs, and using a dynamic adaptation field matching algorithm, a corresponding detection method is matched for each detection item in the optimal detection item set, generating a detection task list.

[0012] Furthermore, the target features include material type, core performance parameters, and historical anomaly rate. Power materials are then classified and detected using a power material clustering classification model, specifically including:

[0013] Multiple target features are acquired and fused into a clustering feature vector;

[0014] The manifold clustering distance is obtained by using a power material clustering classification model and a manifold clustering algorithm.

[0015] Based on the manifold clustering distance, the power materials to be tested are clustered and classified, and the corresponding test classification of the power materials to be tested is given.

[0016] Furthermore, based on manifold clustering distance, the electrical materials to be inspected are clustered and classified, and the corresponding inspection classification of the electrical materials to be inspected is given, specifically including:

[0017] Clustering algorithm is selected based on manifold clustering distance;

[0018] The number of clusters is determined by combining domain knowledge and effectiveness metrics;

[0019] Clustering is performed using the selected clustering algorithm to divide the electrical materials to be detected into several clusters;

[0020] Analyze the feature distribution of each cluster to obtain the risk characteristics of each cluster;

[0021] Based on the risk characteristics of the cluster, risk labels are obtained, and the risk labels and material types are integrated to provide the corresponding detection classification for the power materials to be detected.

[0022] Furthermore, the construction of the detection requirement model specifically includes:

[0023] Obtain historical inspection reports and inspection requirement rule sets for various types of power equipment;

[0024] The historical inspection reports of various types of power materials are analyzed using a semantic tensor extraction model for power materials, and target information is extracted to construct a standardized information database.

[0025] By combining a standardized information database and a set of testing requirements rules, a testing demand model is constructed.

[0026] Furthermore, the target information includes the type of material, testing items, and testing result data;

[0027] By combining a standardized information database and a set of testing requirements rules, a testing requirements model is constructed, which specifically includes:

[0028] Data on material types, testing items, and testing results are obtained from a standardized information database. The set of testing requirement rules is transformed into a rule field to form a structured dataset, which is then divided into a training set and a validation set.

[0029] Obtain the initial detection requirement model, and based on the training set, provide the rule field gravity, historical data potential energy, and rule deviation penalty;

[0030] By combining the gravitational force of the rule field, the potential energy of historical data, and the penalty for rule deviation, the detection requirement field strength for each detection item is obtained;

[0031] Based on the field strength required for detection, a set of candidate detection items is provided;

[0032] The parameters of the initial detection requirement model are optimized through training, and then validated and adjusted using a validation set to obtain the detection requirement model.

[0033] Furthermore, the construction of the detection method library specifically includes:

[0034] The testing method set includes a variety of testing methods and the corresponding testing items for each testing method;

[0035] The detection attribute data includes resource consumption parameters and performance index data. Resource consumption parameters include detection equipment occupancy, manpower requirements, and time costs. Performance index data includes detection accuracy, applicable scenarios, and historical anomaly correlation.

[0036] Dynamic detection adaptation rules include detection priority strategy and detection alternative strategy.

[0037] Furthermore, the construction of the multi-objective optimization function specifically includes:

[0038] Based on a standardized information database and a set of testing requirements rules, a set of quality risks is given, and the coverage of the candidate testing item set to the quality risks is determined.

[0039] By combining the upper limit of detection resources, the fit between the candidate detection item set and the available resources is obtained;

[0040] By combining the time cost ceiling, the time efficiency of the candidate detection item set is obtained;

[0041] By combining coverage, adaptability, and time efficiency, a multi-objective optimization function is constructed, and constraints are set, including mandatory inspection item constraints, resource upper limit constraints, and time upper limit constraints.

[0042] Furthermore, a dynamic fitness field matching algorithm is employed to match a corresponding detection method for each detection item in the optimal detection item set, specifically including:

[0043] For each detection item in the optimal detection item set, obtain the detection performance, matching degree with real-time resources, and task urgency of each detection method associated with the detection item;

[0044] By combining the detection efficiency of each detection method, its matching degree with real-time resources, and the urgency of the task, the dynamic adaptability of the detection method is obtained.

[0045] The dynamic fit of each detection method is compared, and the detection method with the highest dynamic fit is selected as the detection method that matches the current detection item, until the matching of all detection items in the optimal detection item set is completed.

[0046] Furthermore, it also includes:

[0047] Generate a finite number of effective combinations of detection methods based on the detection methods in the detection method library;

[0048] By combining the detection performance, matching degree with real-time resources, task urgency, tunneling probability and combination gain of each combined detection method, the dynamic adaptability of the combined detection method is obtained.

[0049] Compare the dynamic fit of each combined detection method, and select the combined detection method with the highest dynamic fit as the combined detection method that matches the current detection item;

[0050] The dynamic fit of the combined detection method is specifically expressed as follows:

[0051] ;

[0052] Among them, D j Eff(M) represents the dynamic fitness of the j-th combined detection method, where m is the total number of combined detection methods. j ) is a combined detection method M j The detection efficiency of Res(M) j (Z) is a combined detection method M j The degree of matching with real-time resource Z, Urgent(T) is the task urgency, T represents the remaining time of the task, and P is the task urgency. tunnel It is the tunneling probability. It is the combined gain;

[0053] The tunneling probability is specifically expressed as:

[0054] ;

[0055] Where h is a resource adaptation constant, For resource equivalent quality, V barrier D represents the resource barrier, and D represents the resource substitution distance.

[0056] Resource barriers are specifically represented as:

[0057] ;

[0058] in, These are the weighting coefficients; The difference in detection accuracy between the combined detection method and the detection method being replaced. T represents the resource consumption cost ratio of the combined detection method to the detection method being replaced. win Pressure from the time window;

[0059] Combined gain Specifically, it is expressed as follows:

[0060] ;

[0061] Where n is the number of detection methods that make up the combined detection method; Con k Let be the confidence level of the k-th detection method. OP is the execution order penalty factor.

[0062] On the other hand, a power material inspection task generation device, employing the aforementioned power material inspection task generation method, includes:

[0063] The detection clustering and classification module is used to analyze the information of the power materials to be detected, provide multiple target features, process the target features through the power material clustering and classification model, and give the detection classification of the power materials.

[0064] The inspection item acquisition module is used to obtain a set of candidate inspection items by combining the inspection classification of power materials and the pre-built inspection requirement model.

[0065] The detection item optimization module is used to provide the optimal set of detection items based on a pre-established detection method library, using a multi-objective optimization function and iterative optimization through a genetic algorithm.

[0066] The detection method matching module combines real-time detection resources and time costs, and uses a dynamic adaptation field matching algorithm to match the corresponding detection method for each detection item in the optimal detection item set, generating a detection task list.

[0067] The present invention provides a task generation method and apparatus for power material inspection, which has at least the following beneficial effects:

[0068] (1) This invention, by constructing a semantic tensor extraction model for power materials, achieves in-depth analysis of historical inspection reports and standardized data storage, improving the efficiency and accuracy of data processing. Through semantic enhancement tensors and contextual attention mechanisms, it accurately extracts key information on material types, inspection items, and results, providing a data foundation for the construction of subsequent inspection demand models. In addition, the inspection demand model, through the comprehensive calculation of rule field attraction, historical data potential energy, and rule deviation penalties, achieves accurate prediction and dynamic adjustment of inspection demands, effectively avoiding over-inspection or under-inspection, and improving the scientific nature and pertinence of inspection work.

[0069] (2) This invention achieves refined management of power materials by selecting material type, core performance parameters, and historical anomaly rate as feature vectors and using manifold clustering algorithm for quantitative classification. Regarding the optimization of detection items, this method aims at comprehensiveness, resource adaptability, and time efficiency, constructing a multi-objective optimization function and using a genetic algorithm for iterative optimization to output the optimal set of detection items. This not only ensures the coverage and effectiveness of detection items but also improves the utilization efficiency of detection resources and reduces detection costs. The introduction of a dynamic adaptability field detection method matching algorithm further enhances the matching accuracy between detection methods and resources, providing a strong guarantee for the safe operation of power materials.

[0070] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0071] Figure 1 A flowchart of a method for generating power material inspection tasks provided by the present invention;

[0072] Figure 2 This is a flowchart illustrating the power material detection and classification using a power material clustering classification model according to a certain embodiment of the present invention.

[0073] Figure 3 This is a diagram of the power material clustering and classification model architecture according to a certain embodiment of the present invention;

[0074] Figure 4 This is a flowchart illustrating a detection and classification process based on manifold clustering, according to a certain embodiment of the present invention.

[0075] Figure 5 This is a diagram of the semantic tensor extraction model for power materials according to a certain embodiment of the present invention.

[0076] Figure 6 This is a flowchart illustrating the construction of a detection requirement model according to a certain embodiment of the present invention.

[0077] Figure 7 This is a flowchart illustrating the construction of a multi-objective optimization function according to a certain embodiment of the present invention;

[0078] Figure 8 This is a schematic diagram of a power material inspection task generation device module provided by the present invention. Detailed Implementation

[0079] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0080] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0081] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0082] like Figure 1 As shown, the present invention provides a method for generating power material inspection tasks, the specific steps of which include:

[0083] Based on the analysis of information on power materials to be detected, multiple target features are given. The target features are then processed through a power material clustering classification model to provide a detection classification of the power materials.

[0084] By combining the inspection classification of power materials with the pre-built inspection demand model, a set of candidate inspection items is obtained;

[0085] Based on a pre-established detection method library, an optimal set of detection items is given by using a multi-objective optimization function and iterative optimization through a genetic algorithm.

[0086] By combining real-time detection resources and time costs, and using a dynamic adaptation field matching algorithm, a corresponding detection method is matched for each detection item in the optimal detection item set, generating a detection task list.

[0087] The system analyzes information on power materials to be inspected, extracting target features such as material type, core performance parameters, and historical anomaly rates. A power material clustering classification model is then used to process these target features, achieving accurate classification of the power materials under inspection. The materials are divided into inspection categories with different risk levels, and the classification results more closely reflect the current actual state of the materials. A multi-objective optimization function is constructed to balance inspection comprehensiveness, resource adaptability, and time efficiency, efficiently and rationally selecting the necessary inspection items. A dynamic adaptability field matching algorithm adapts to real-time resource changes, optimizes resource utilization, and quickly and efficiently matches the optimal inspection method for each inspection item.

[0088] Based on the analysis of information on the power materials to be inspected, several target features are given, including:

[0089] Collect multi-source information on the power materials to be inspected, including basic attributes (such as material type, rated parameters, and years of operation), historical inspection data (such as historical inspection items, inspection results, and historical anomaly rates), and standard requirements (such as mandatory inspection items and risk association rules in power industry standards).

[0090] Select features that reflect material risks and meet standard requirements from multi-source information to avoid redundancy, such as selecting risk-related features, type attribute-related features, and standard requirement-related features as target features;

[0091] The selected target features are used as input to the power material clustering and classification model.

[0092] like Figure 2 As shown, the target features are processed using a power material clustering classification model to provide the detection and classification of power materials, specifically including:

[0093] Multiple target features are acquired and fused into a clustering feature vector;

[0094] The manifold clustering distance is obtained by using a power material clustering classification model and a manifold clustering algorithm.

[0095] Based on the manifold clustering distance, the power materials to be tested are clustered and classified, and the corresponding test classification of the power materials to be tested is given.

[0096] In this example, when classifying the electrical materials under test, the electrical material clustering classification model used is a deep learning-enhanced manifold clustering model, such as... Figure 3 As shown, it includes a feature preprocessing layer, a feature fusion layer, a manifold embedding layer, a manifold distance calculation layer, and a clustering and classification layer:

[0097] The feature preprocessing layer standardizes and normalizes target features such as material type, core performance parameters and historical detection anomaly rate, such as encoding the material type, standardizing the rated voltage value in the core performance parameters, and normalizing the historical detection anomaly rate.

[0098] The feature fusion layer concatenates the preprocessed target features into a clustering feature vector;

[0099] The manifold embedding layer uses the isometric mapping algorithm (ISOMAP) or the local linear embedding algorithm (LLE) to map the concatenated high-dimensional clustering feature vectors to the low-dimensional manifold space. This can preserve the non-linear structure of high-dimensional data and avoid the loss of key information when mapping high-dimensional data to low-dimensional space.

[0100] The manifold distance calculation layer calculates the Riemannian metric in the manifold space and integrates the risk correction term to obtain the final manifold clustering distance. The Riemannian metric reflects the original geometric differences of the features, and the risk correction term is used to control the degree of influence of historical anomalies on the manifold clustering distance, making it easier for high-risk materials (such as cables with high historical anomaly rates) to be clustered into one class.

[0101] The clustering classification layer uses hierarchical clustering or K-means clustering to classify the electrical materials under test based on the final manifold clustering distance, providing a basis for subsequent acquisition and optimization of testing items (such as adding testing items for high-risk categories).

[0102] Furthermore, the specific expression for obtaining the manifold clustering distance is:

[0103] ;

[0104] Where, d pq Metric(F) is the clustering distance between the manifolds of materials p and q, where M is the characteristic manifold space of the power materials. p F q F is the clustering feature vector F in the characteristic manifold space of power materials. p With F q Riemannian metric It is a risk correction item. Risk(p, q) is the risk sensitivity coefficient, and Risk(p, q) is the historical abnormal correlation between material p and q.

[0105] Understandably, the manifold space of power material features is a low-dimensional curved space mapped by high-dimensional clustering feature vectors, learned from high-dimensional clustering feature vectors through manifold learning algorithms such as ISOMAP and LLE. The Riemann metric measures the similarity between clustering feature vectors; it is the shortest path length between clustering feature vectors in the manifold space. Its calculation depends on the geometric structure of the manifold space, and common calculation methods include geodesic distance, local linear reconstruction distance, and probability distribution distance. The risk correction term consists of two parts: a risk sensitivity coefficient and historical anomaly correlation. It is used to adjust the manifold clustering distance between materials, ensuring that materials with high historical anomaly correlation (such as transformers that have both experienced "oil quality deterioration" or "partial discharge" anomalies) are clustered together. Smaller distances make them more likely to be classified into the same category; the risk sensitivity coefficient can be set based on domain knowledge or adjusted experimentally to control the intensity of risk correction; the historical anomaly correlation degree is used to measure the degree of correlation between anomalies that occurred in the historical testing process of different materials, reflecting the similarity between materials in anomaly type (such as "oil quality deterioration" or "winding insulation reduction") or anomaly frequency. It is calculated from historical testing data in a standardized information database. The calculation method can be "overlap of historical anomaly types of two materials × weighted sum of anomaly frequency" (e.g., if the overlap of the "partial discharge" anomaly type of materials p and q is 0.8 and the weighted sum of anomaly frequency is 0.9, then Risk(p, q) = 0.8 * 0.9 = 0.72).

[0106] In one implementation, such as Figure 4As shown, based on manifold clustering distance, the electrical materials to be tested are clustered and classified, and the corresponding detection classification of the electrical materials to be tested is given, specifically including:

[0107] Clustering algorithm is selected based on manifold clustering distance;

[0108] The number of clusters is determined by combining domain knowledge and effectiveness metrics;

[0109] Clustering is performed using the selected clustering algorithm to divide the electrical materials to be detected into several clusters;

[0110] Analyze the feature distribution of each cluster to obtain the risk characteristics of each cluster;

[0111] Based on the risk characteristics of the cluster, risk labels are obtained, and the risk labels and material types are integrated to provide the corresponding detection classification for the power materials to be detected.

[0112] Specifically, manifold clustering distance integrates feature similarity (Riemannian metric) and risk correlation (historical anomaly correlation), and selects algorithms that can handle arbitrary distance metrics, such as hierarchical clustering and spectral clustering.

[0113] The number of clusters is determined by combining domain knowledge (such as the "risk level classification" of power operation and maintenance) and clustering effectiveness indicators (such as silhouette coefficient, elbow method, etc.).

[0114] Using a selected clustering algorithm and manifold clustering distance as a metric, the electrical materials to be detected are merged into several clusters. Taking hierarchical clustering as an example, this process specifically includes:

[0115] (1) Initialize each power material to be detected as an independent cluster;

[0116] (2) Obtain the average manifold clustering distance between all points in cluster A and all points in cluster B, and use it as the inter-cluster distance;

[0117] (3) Each time, the two clusters with the smallest inter-cluster distance are merged (e.g., the manifold clustering distance between "oil-immersed transformers that have been in operation for 10 years" and "oil-immersed transformers that have been in operation for 8 years" is the smallest, so they are merged into one cluster).

[0118] (4) Repeat steps (2) and (3) until the number of clusters is reduced to the determined number of clusters;

[0119] Statistically analyze the characteristic distribution of each cluster (such as material type, historical anomaly rate, years of operation, etc.), understand the risk meaning of the cluster, and obtain the risk characteristics of the cluster (such as historical anomaly rate, operating status, etc.).

[0120] Based on the risk characteristics of the clusters, each cluster is assigned a risk label (such as high risk, normal, low risk). The risk label is then integrated with the material type to provide the corresponding detection classification for the power materials to be tested.

[0121] In one implementation method, the construction of the detection requirement model specifically includes:

[0122] Obtain historical inspection reports and inspection requirement rule sets for various types of power equipment;

[0123] The historical inspection reports of various types of power materials are analyzed using a semantic tensor extraction model for power materials, and target information is extracted to construct a standardized information database.

[0124] By combining a standardized information database and a set of testing requirements rules, a testing demand model is constructed.

[0125] The set of testing requirements rules includes:

[0126] List of mandatory inspection items for various types of power equipment, such as the mandatory inspection items for 10kV cross-linked polyethylene cables, which include insulation resistance test and withstand voltage test;

[0127] The threshold values ​​for the test indicators of each mandatory test item include the pass threshold and the warning threshold. For example, the pass threshold for cable insulation resistance test is ≥1000MΩ, and the warning threshold is ≥800MΩ.

[0128] The testing cycles for different types of power equipment in different scenarios include the pre-commissioning testing cycle, the daily operation testing cycle, and the post-fault repair testing cycle, to ensure the timeliness of testing and avoid "over-testing" or "missed testing".

[0129] The rules for linking testing items are as follows: when the test result of a certain item reaches a specific condition, other testing items must be carried out in conjunction. For example, if the "insulation resistance test result of a certain cable is < warning threshold (800MΩ)", then "partial discharge detection" will be triggered to further investigate insulation defects.

[0130] Special testing requirements are set for specific types of power equipment, such as adding "temperature rise monitoring" to high-load (voltage ≥ 90% of rated voltage) transformers to prevent overheating failures.

[0131] The detection requirement rule set provides rule constraints for the detection requirement model, provides constraints for the optimization of detection projects, and provides selection criteria for matching detection methods.

[0132] In this example, the semantic tensor extraction model for power materials can be obtained through pre-training by integrating power knowledge graphs and neural network models such as BERT. Figure 5As shown, the semantic tensor extraction model for power materials includes a word embedding layer, a semantic enhancement layer, a contextual attention layer, a professional feature normalization layer, and a semantic fusion layer. For historical inspection reports, preprocessing is performed first, then word segmentation is used to generate a word embedding matrix, followed by semantic enhancement using a knowledge graph, obtaining contextual attention weights, feature extraction and normalization, and finally, fusing this information to obtain the semantic vector for each report. Subsequently, target information such as material type, inspection item, and inspection result are extracted from the semantic vector.

[0133] The process of parsing historical inspection reports using the semantic tensor extraction model for power materials includes:

[0134] The word embedding layer takes the word sequence after segmentation of the historical detection report as input, and uses a pre-trained word embedding model (such as Word2Vec, BERT) to convert the word sequence into a low-dimensional vector and generate a word embedding matrix. In this way, the unstructured text words are converted into computable semantic vectors.

[0135] The semantic enhancement layer takes the word embedding matrix as input, queries the triples associated with the words in the pre-constructed power knowledge graph, calculates the semantic enhancement tensor of the words, and enhances the semantics of the words in the power domain.

[0136] The contextual attention layer takes the current sentence topic and the global detection context as input, uses an attention mechanism to calculate the sentence weight, highlights key sentences related to detection in historical detection reports, and filters out irrelevant information (such as report headings, compilers, etc.).

[0137] The professional feature normalization layer takes the power industry features in the input sentence, extracts the feature values ​​and units through regular expressions or rule engines, and then normalizes them into a standardized format. This transforms unstructured detection results (such as natural language descriptions) into computable structured data, laying the foundation for the subsequent construction of a standardized information database.

[0138] The semantic fusion layer merges the outputs of the above layers into a report-level semantic extraction vector, which is a global semantic representation of the historical detection report and contains all key information such as "material type", "detection item" and "detection result".

[0139] Specifically, in this example, the semantic extraction vector for electrical materials is represented as follows:

[0140] ;

[0141] Among them, S i E is the report-level semantic extraction vector of the i-th detection report, where n is the total number of sentences in the i-th report. ij It is the word embedding matrix of the j-th sentence in the i-th report, KG(W ij ) is the power knowledge graph for the vocabulary W ijThe semantic enhancement tensor is computed using knowledge graph triples, Attn(T) ij C j ) is a context-based attention weight, T ij C is the topic of the current sentence. j For the global detection context, Norm(F) ij F is the power industry characteristic normalization function. ij Let be the professional feature vector of the j-th sentence.

[0142] Because the report-level semantic extraction vector incorporates all key information, the target information can then be extracted using classification models or entity extraction models, such as:

[0143] For each type of material, a text classification model (such as CNN, BERT, etc.) is used to classify the semantic extraction vector at the report level and output the material type.

[0144] For the detected items, the Named Entity Recognition (NER) model is used to extract them from the report-level semantic extraction vector;

[0145] For the detection results, a rule engine or sequence labeling model is used to extract the semantic vector from the report level.

[0146] The aforementioned target information is collected to construct a standardized information database.

[0147] This example uses a hierarchical fusion architecture of the power material semantic tensor extraction model to achieve in-depth analysis and target information extraction of historical inspection reports, and finally builds a standardized information database, providing reliable data support for the construction of inspection requirement models.

[0148] Furthermore, such as Figure 6 As shown, by combining a standardized information database and a set of testing requirement rules, a testing requirement model is constructed, which specifically includes:

[0149] Data on material types, testing items, and testing results are obtained from a standardized information database. The set of testing requirement rules is transformed into a rule field to form a structured dataset, which is then divided into a training set and a validation set.

[0150] Obtain the initial detection requirement model, and based on the training set, provide the rule field gravity, historical data potential energy, and rule deviation penalty;

[0151] By combining the gravitational force of the rule field, the potential energy of historical data, and the penalty for rule deviation, the detection requirement field strength for each detection item is obtained;

[0152] Based on the field strength required for detection, a set of candidate detection items is provided;

[0153] The parameters of the initial detection requirement model are optimized through training, and then validated and adjusted using a validation set to obtain the detection requirement model.

[0154] In practice, the process of obtaining structured datasets includes:

[0155] The unstructured rules of the detection requirement rule set are converted into a quantifiable rule field matrix. For example, mandatory items are labeled with binary symbols (1 = mandatory, 0 = non-mandatory), detection thresholds and detection cycles are represented numerically, and association rules are represented by conditional weights to indicate the probability of associated items.

[0156] The material data (material type, testing items and testing results data) in the standardized information database are associated with the rule field matrix to form a structured dataset, which is then divided into training set and validation set in a ratio of 7:3 or 8:2 for model training and evaluation.

[0157] The initial detection requirement model can be a fusion model based on neural networks, including an input layer, a feature calculation layer, a hidden layer, and an output layer:

[0158] The input layer is used to receive material status characteristics (such as material type and core performance parameters), historical detection data (such as historical anomaly rate and fault records), and rule matching data (such as the deviation between the current detection result and the standard rule).

[0159] The feature calculation layer is used to obtain the gravity of the rule field, the potential energy of historical data, and the rule deviation penalty;

[0160] The hidden layer is used to fuse the gravity of the rule field, the potential energy of historical data, and the rule deviation penalty to obtain the field strength required for detection;

[0161] The output layer generates a set of candidate detection items based on the required field strength.

[0162] The gravitational pull of a defined field represents the degree of fit between the current state of electrical equipment (e.g., an "oil-immersed transformer") and a defined field (e.g., mandatory oil quality testing). A higher degree of fit indicates that the equipment better meets the regulatory requirements and that the testing needs are more clearly defined. Specifically, this is expressed as:

[0163] ;

[0164] In the formula, It is the gravitational field of the k-th type of material, S ck Characteristics of the material status. To detect the rule-based field mapping of the requirement set; This is the matching degree function used to calculate S. ck and Similarity; W g Let be the gravimetric submatrix, representing the correlation weights between rules.

[0165] Understandably, the material status characteristics come from a standardized information database, such as "oil-immersed transformer" and "rated capacity 500kVA"; the rule field mapping is the quantitative result of the rule set required for detection; the matching degree function can take the form of Boolean matching, calculating weighted similarity, or it can be the matching probability output by the model through training; the graviton matrix is ​​the correlation strength matrix between each rule in the rule field.

[0166] Historical data potential energy represents the degree of risk accumulation from historical anomalies in power equipment data. The more historical anomalies, the higher the risk potential energy, and the more urgent the need for detection. Specifically, this is expressed as follows:

[0167] ;

[0168] In the formula, It is the potential energy of historical data, I c This refers to an unusual characteristic of a certain material. This refers to the historical anomalous potential energy field of similar materials; Density (I c () represents the feature density, indicating the frequency of occurrence of a certain anomalous feature; The distance is the manifold distance, representing the similarity between the current material characteristics and historical anomalous potential energy fields. The smaller the distance, the higher the risk.

[0169] It is understandable that the abnormal features come from a standardized information database, such as "historical abnormality rate 5%" and "number of failures 3 times"; the historical abnormal potential energy field of similar materials comes from the historical abnormal data in the standardized information database, such as "the distribution of the abnormality rate of oil-immersed transformers in the past 5 years".

[0170] Rule deviation penalty indicates the degree of deviation between the current detection result of electrical equipment and the rule threshold. The greater the deviation, the higher the penalty, and the more urgent the need for detection. Specifically:

[0171] ;

[0172] In the formula, It is a rule deviation penalty, meaning that a penalty is triggered when the rule deviates from the boundary. R ck The current rule matching degree, For standard rule boundaries; The tolerance range indicates the allowable deviation range of the rule; It is the rule penalty coefficient, which is determined by the importance of the rule.

[0173] Understandably, the current rule matching degree comes from the current test results in the standardized information database, such as "oil quality test result 0.05mg / L"; the standard rule boundary comes from the threshold of the test requirement rule set, such as "pass threshold 0.1mg / L".

[0174] The field strength for testing demand is a weighted sum of the gravitational force of the rule-defined field, the potential energy of historical data, and the penalty for rule deviation. It represents the urgency of the testing demand for a certain test item on electrical materials. The higher the field strength, the more urgent the testing demand. Specifically, it is expressed as follows:

[0175] ;

[0176] Among them, D k The testing demand field strength for category k materials. These are weighting coefficients, which can be set flexibly.

[0177] Based on the required field strength for detection, a set of candidate detection items is generated, specifically including:

[0178] Obtain a list of mandatory inspection items for this type of material from the set of inspection requirements rules (e.g., "mandatory inspection items for oil-immersed transformers: oil quality testing, winding insulation testing"), and include them in the candidate inspection item set;

[0179] Based on the set detection requirement field strength threshold, detection items whose detection requirement field strength exceeds the threshold are selected and included in the candidate detection item set;

[0180] Based on the association rules in the set of testing requirements (such as "when the test result of a certain testing project reaches a certain condition, other projects need to be linked"), when the testing demand field strength of a certain testing project is too high, add linked testing projects and include them in the candidate testing project set.

[0181] The duplicate detection items in the candidate detection item set are deduplicated to obtain the final candidate detection item set.

[0182] Understandably, based on the set detection requirement field strength threshold, the detection requirement field strength can be divided into "high field strength", "medium field strength" and "low field strength". The threshold can be an empirical value or an experimental value.

[0183] By using training data, the model parameters are optimized to match the predicted detection demand field strength with the actual detection needs (such as historical detection items for this type of material). Then, the model performance is evaluated using validation data, comparing the predicted detection needs with actual needs and calculating metrics such as accuracy, recall, and F1-score. Based on the feedback from these metrics, the model parameters are adjusted to improve accuracy and generalization ability. After training and validation, an optimized detection demand model is obtained. The input to this model is material data from a standardized information database (e.g., "oil-immersed transformer, oil quality test result 0.05mg / L") and the rule field of the detection requirement rule set (e.g., "mandatory inspection mark 1, threshold 0.1mg / L"). The output is the detection demand field strength and a set of candidate detection items.

[0184] The demand detection model constructed in this example ensures that no mandatory items are missed through the attraction of the rule field, balances the detection frequency by using the potential energy of historical data and the penalty for rule deviation, and dynamically adjusts candidate items by associating rules and material status. This achieves accurate prediction of detection demand, and the resulting set of candidate detection items forms the basis for subsequent optimization of detection items.

[0185] In one implementation, the construction of the detection method library includes:

[0186] The testing method set includes a variety of testing methods and the corresponding testing items for each testing method;

[0187] The detection attribute data includes resource consumption parameters and performance index data. Resource consumption parameters include detection equipment occupancy, manpower requirements, and time costs. Performance index data includes detection accuracy, applicable scenarios, and historical anomaly correlation.

[0188] Dynamic detection adaptation rules include detection priority strategy and detection alternative strategy.

[0189] In this example, the detection method library associates at least one detection method with each detection item, adapting to different scenario requirements and avoiding the limitations that may exist with a single method; the quantification of resource consumption parameters and performance indicators makes resource requirements calculable and resource conflicts can be warned, improving detection accuracy and targeting; the setting of dynamic detection adaptation rules can cope with resource fluctuations and emergencies, enhancing detection flexibility.

[0190] In one implementation, such as Figure 7 As shown, the construction of the multi-objective optimization function specifically includes:

[0191] Based on a standardized information database and a set of testing requirements rules, a set of quality risks is given, and the coverage of the candidate testing item set to the quality risks is determined.

[0192] By combining the upper limit of detection resources, the fit between the candidate detection item set and the available resources is obtained;

[0193] By combining the time cost ceiling, the time efficiency of the candidate detection item set is obtained;

[0194] By combining coverage, adaptability, and time efficiency, a multi-objective optimization function is constructed, and constraints are set, including mandatory inspection item constraints, resource upper limit constraints, and time upper limit constraints.

[0195] It is understandable that historical anomaly data from historical test reports in the standardized information database, mandatory test items in the test requirement rule set, and risk association rules are extracted to form a quality risk set.

[0196] Coverage is the proportion of the candidate detection item set that covers the quality risk set, specifically expressed as:

[0197] ;

[0198] In the formula, Coverage(X) r , R r X is the set of candidate detection items. r Quality risk R r The coverage; n is the total number of risks; It is the weight of the k-th quality risk, which is determined by the severity of the risk and can be an empirical value or an experimental value. Indicates the indicator function, if the detection item Coverage risk r k ,but ,otherwise ; This represents the weighted sum of risks for the set of candidate detection items; This represents the total weighted risk of the quality risk set.

[0199] Fit is used to quantify the degree of matching between the resource consumption (equipment, manpower) of a candidate detection item set and the available resources. The higher the fit, the higher the resource utilization rate. Specifically, it is expressed as:

[0200] ;

[0201] In the formula, Resource(X) r R) is the set of candidate detection items X. r The degree of fit with available resources R; The total number of test items in the candidate test item set; Indicates the testing items Resource consumption; R represents the total resource consumption of the candidate detection item set; total This represents the maximum available resources.

[0202] Time efficiency is used to quantify the degree to which the time consumption of the candidate detection item set matches the time limit; the higher the efficiency, the less time is consumed. Specifically, it is expressed as:

[0203] ;

[0204] In the formula, Time(X) r (T) is the set of candidate detection items X. r Time efficiency; The total number of test items in the candidate test item set; This is the lth test item Time cost; T represents the total time cost of the candidate detection set; max This is the upper limit of the time limit.

[0205] By weighting coverage, fit, and time efficiency, a target function is formed for easy optimization. The target weights are adjusted to reflect the importance of different factors, and constraints are set, specifically as follows:

[0206] ;

[0207] ;

[0208] Where F represents a multi-objective function, It is the target weight, and , B r This is the set of mandatory inspection items for Category r materials.

[0209] In this example, a multi-objective optimization function is constructed and constraints are set. "Coverage" ensures that the detection items cover the main quality risks, "fitness" and "time efficiency" ensure that the detection items do not exceed the resource and time limits, and the importance of the objectives is adjusted by weights, thus balancing the comprehensiveness of detection with resource and time efficiency.

[0210] After constructing multiple objective optimization functions and setting constraints, a genetic algorithm is used for iterative optimization to obtain the optimal set of detection items. The specific process includes:

[0211] (1) The candidate detection item set is represented by a binary string. Each binary bit corresponds to a detection item. 1 indicates that the item is selected and 0 indicates that it is not selected. The binary bit corresponding to the mandatory item is fixed at 1 to ensure that no mandatory item is missed. The gene bit corresponding to the non-mandatory item is randomly 0 or 1 to allow for flexible selection.

[0212] (2) Randomly generate a certain number of binary strings that conform to the encoding rules of step (1) to represent the candidate item set.

[0213] (3) For each set of candidate items, calculate the multi-objective optimization function value and remove the set of candidate items that violate the constraints.

[0214] (4) Use the competition selection method or other selection methods to select a certain number of candidate projects with high multi-objective optimization function values.

[0215] (5) Cross the candidate item set in step (4), that is, randomly select several different candidate item sets, swap the same position of the binary string corresponding to the different candidate item sets, generate a new binary string, that is, a new candidate item set, and remove the candidate item set that violates the constraint conditions.

[0216] (6) Mutate the candidate item set obtained in step (5), that is, randomly select several candidate item sets, and randomly select the binary bit corresponding to a non-mandatory item, and flip (1→0, 0→1) the binary bit. Eliminate the candidate item set that violates the constraints.

[0217] (7) Repeat steps (3) to (6) until the maximum number of iterations or the maximum value of the multi-objective optimization function changes less than a certain threshold (convergence).

[0218] (8) Output the set of candidate items with the largest multi-objective optimization function value as the optimal detection item set.

[0219] In this example, the optimal set of detection items obtained through iterative optimization using a genetic algorithm covers the main quality risks, exhibiting high resource adaptability and time efficiency. The entire process balances detection comprehensiveness, resource adaptability, and time efficiency, solving the problems of "high subjectivity, resource waste, and insufficient coverage" in traditional detection item selection, and achieving accurate, efficient, and executable detection tasks.

[0220] In one implementation, a dynamic fitness field matching algorithm is used to match a corresponding detection method for each detection item in the optimal detection item set, specifically including:

[0221] For each detection item in the optimal detection item set, obtain the detection performance, matching degree with real-time resources, and task urgency of each detection method associated with the detection item;

[0222] By combining the detection efficiency of each detection method, its matching degree with real-time resources, and the urgency of the task, the dynamic adaptability of the detection method is obtained.

[0223] The dynamic fit of each detection method is compared, and the detection method with the highest dynamic fit is selected as the detection method that matches the current detection item, until the matching of all detection items in the optimal detection item set is completed.

[0224] Understandably, the detection method library contains the detection methods associated with each detection item, as well as the performance indicators, resource consumption, and dynamic adaptation rules for each detection method. Real-time resource data can be obtained in real time through dynamic resource detection technologies and other means.

[0225] For each test item in the optimal test item set (e.g., "insulation resistance test" or "resistance voltage test"), retrieve all associated test methods from the test method library, and obtain three key parameters for each test method: test efficiency, resource matching degree, and task urgency. Test efficiency refers to the performance of the test method (e.g., test accuracy and applicability to various scenarios); the higher the value, the more effective the method. Resource matching degree is the degree of matching between the test method's resource consumption (equipment, manpower, time) and real-time resources; the higher the value, the higher the resource utilization rate. Task urgency is the urgency of the test task (the less time remaining, the higher the urgency); the higher the value, the greater the weight of the method's "time efficiency".

[0226] In this example, no specific restrictions are placed on the calculation methods for detection efficiency, resource matching degree, and task urgency.

[0227] By combining the detection performance of each detection method, its compatibility with real-time resources, and the urgency of the task, the dynamic adaptability of the detection method is obtained. Specifically, this is expressed as:

[0228] ;

[0229] Among them, D j M is the dynamic fitness of the j-th detection method. j This is the j-th detection method, where m is the total number of detection methods associated with the detection item, and Eff(M) j ) represents the detection efficiency of the j-th detection method, Res(M) j ,Z) is the matching degree between the j-th detection method and the real-time resource Z, Urgent(T) is the task urgency, and T represents the remaining time of the task.

[0230] After obtaining the dynamic fit of all detection methods associated with the current detection project, the detection method with the highest dynamic fit is selected as the detection method that matches the current detection project.

[0231] The following is an example of matching the detection method for each detection item in the optimal set of detection items ("oil quality detection", "winding insulation resistance test") using the dynamic fit field matching algorithm:

[0232] For the "oil quality testing" item, it is associated with "spectral analysis" and "chromatographic analysis". The detection efficiency, matching degree with real-time resources, and task urgency of the two detection methods are calculated respectively, and the dynamic adaptability is obtained for each. Assuming that the dynamic adaptability is 0.85 and 0.78 respectively, the "spectral analysis" method with the largest dynamic adaptability of 0.85 is selected. For the "winding insulation resistance testing" item, it is associated with "megohmmeter method" and "insulation resistance tester method". The detection efficiency, matching degree with real-time resources, and task urgency of the two detection methods are calculated respectively, and the dynamic adaptability is obtained for each. Assuming that the dynamic adaptability is 0.90 and 0.82 respectively, the "megohmmeter method" with the largest dynamic adaptability of 0.90 is selected.

[0233] After matching the testing methods for all testing items in the optimal testing item set, supplement the execution information (such as testing personnel grouping, equipment usage time period, testing order, etc.) to generate the final testing task list.

[0234] This method organically combines data, standards, and resources through core models and algorithms, avoiding the drawbacks of manually defining tasks and optimizing the entire process of detection tasks from data to execution, thereby improving the accuracy and efficiency of detection.

[0235] In another implementation, when a resource conflict is detected, a combined detection method can be generated and its dynamic adaptability can be obtained. By comparing the magnitudes of the dynamic adaptability, a combined detection method is selected, specifically including:

[0236] (1) Genetic algorithms or other heuristic search algorithms are used to generate a finite number of effective combined detection methods based on the detection methods in the detection method library. The combined detection method is composed of multiple detection methods from the detection method library according to a specific logical relationship, and inherits their resource consumption parameters and performance index data.

[0237] If the ultrasonic instrument required for the "partial discharge detection" method using the "ultrasonic testing method" is occupied and will be unavailable for the next X hours, generate a combined detection method that first performs high-frequency current transformer (HFCT) detection, then performs transient ground voltage (TEV) detection, to replace the "ultrasonic testing method".

[0238] (2) Obtain the dynamic fitness of each combined detection method. The dynamic fitness is specifically expressed as follows:

[0239] ;

[0240] Among them, D j Eff(M) represents the dynamic fitness of the j-th combined detection method, where m is the total number of combined detection methods. j ) is a combined detection method M j The detection efficiency of Res(M) j(Z) is a combined detection method M j The degree of matching with real-time resource Z, Urgent(T) is the task urgency, T represents the remaining time of the task, and P is the task urgency. tunnel It is the tunneling probability. It is the combined gain.

[0241] Tunneling probability P tunnel This represents the probability that, in the event of a resource conflict, an alternative solution (a combined detection method) will replace the original solution (a detection method from the detection method library) to overcome the resource barrier. This probability is obtained based on the resource fit constant, resource equivalent quality, resource barrier, and resource substitution distance. Specifically, it is expressed as:

[0242] ;

[0243] Where h is a resource adaptation constant, which can be an empirical value; in this example, it is taken as 0.2~0.5. The equivalent quality of resources is determined by the scarcity of equipment and the redundancy of manpower; V barrier D represents the resource barrier; D is the resource substitution distance, which indicates the degree of difference between the substitute resource and the original resource (values ​​from 0 to 1). The greater the difference, the larger D is.

[0244] Resource barriers V barrier The calculation is based on the detection accuracy deviation, resource consumption cost ratio, and time window pressure between the combined detection method and the replaced detection method. The formula is as follows:

[0245] ;

[0246] in, These are the weighting coefficients; The difference in detection accuracy between the combined detection method and the detection method being replaced. T represents the resource consumption cost ratio of the combined detection method to the detection method being replaced. win The time window pressure is determined by the ratio of the remaining time of the task to the required time.

[0247] Combined gain This is used to compensate for the confidence loss caused by substitution in the combined detection method, and is obtained based on the confidence and order penalty of each detection method that makes up the combined detection method. Its calculation formula is:

[0248] ;

[0249] Where n is the number of detection methods that make up the combined detection method; Con k Let be the confidence level of the k-th detection method; The order penalty factor and OP (operational order penalty) can be flexibly set based on experience or experimentation. The confidence of the combined detection method is derived from the product of the confidence of each detection method, which is usually lower than that of a single detection method. Sequential execution introduces additional time penalties (the more steps, the higher the probability of error), so it is adjusted through exponential decay. The confidence of the detection method can be obtained from a detection method library. For example, the performance index data (detection accuracy, applicable scenarios, historical anomaly correlation, etc.) corresponding to the detection method can be normalized and weighted to obtain the confidence, which is used to quantify the reliability of the detection method.

[0250] (3) Compare the dynamic fit of each combination detection method and select the combination detection method with the largest dynamic fit as the combination detection method that matches the current detection item.

[0251] In this example, by generating a combined detection method, resource conflicts are transformed into executable alternatives, avoiding long waiting times for detection tasks due to insufficient detection resources and ensuring the timely completion of urgent tasks. Alternative resources are obtained through tunneling probability for alternative detection, and the confidence loss is compensated for by combined gain, balancing detection accuracy, reliability, and cost-effectiveness.

[0252] On the other hand, such as Figure 8 As shown, the present invention provides a power material inspection task generation device, which employs the above-mentioned power material inspection task generation method. The device includes:

[0253] The detection clustering and classification module is used to analyze the information of the power materials to be detected, provide multiple target features, process the target features through the power material clustering and classification model, and give the detection classification of the power materials.

[0254] The inspection item acquisition module is used to obtain a set of candidate inspection items by combining the inspection classification of power materials and the pre-built inspection requirement model.

[0255] The detection item optimization module is used to provide the optimal set of detection items based on a pre-established detection method library, using a multi-objective optimization function and iterative optimization through a genetic algorithm.

[0256] The detection method matching module combines real-time detection resources and time costs, and uses a dynamic adaptation field matching algorithm to match the corresponding detection method for each detection item in the optimal detection item set, generating a detection task list.

[0257] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for generating power material inspection tasks, characterized in that, The specific steps include: Based on the analysis of information on power materials to be detected, multiple target features are given. The target features are then processed through a power material clustering classification model to provide a detection classification of the power materials. By combining the inspection classification of power materials with the pre-built inspection demand model, a set of candidate inspection items is obtained; Based on a pre-established detection method library, an optimal set of detection items is given by using a multi-objective optimization function and iterative optimization through a genetic algorithm. By combining real-time detection resources and time costs, and employing a dynamic fit field matching algorithm, a corresponding detection method is matched for each detection item in the optimal detection item set. Specifically, this includes: generating a finite number of effective combined detection methods based on the detection methods in the detection method library; obtaining the dynamic fit of each combined detection method by combining its detection performance, matching degree with real-time resources, task urgency, tunneling probability, and combined gain; comparing the dynamic fit of each combined detection method, and selecting the combined detection method with the highest dynamic fit as the combined detection method matched with the current detection item. The dynamic fit of the combined detection method is specifically expressed as follows: ； Among them, D j Eff(M) represents the dynamic fitness of the j-th combined detection method, where m is the total number of combined detection methods. j ) is a combined detection method M j The detection efficiency of Res(M) j (Z) is a combined detection method M j The matching degree with real-time resource Z, Urgent(T) is the task urgency, T represents the remaining time of the task, and P is the urgency of the task. tunnel It is the tunneling probability. It is the combined gain, and h is the resource adaptation constant. For resource equivalent quality, V barrier Where D represents the resource barrier and D represents the distance to resource substitution. These are the weighting coefficients. The difference in detection accuracy between the combined detection method and the detection method being replaced. T represents the resource consumption cost ratio of the combined detection method to the detection method being replaced. win Let n be the time window pressure, and n be the number of detection methods that make up the combined detection method. k Let be the confidence level of the k-th detection method. OP is the execution order penalty factor; Generate a list of detection tasks.

2. The method for generating power material inspection tasks according to claim 1, characterized in that, The target features include material type, core performance parameters, and historical anomaly rate. These features are processed using a power material clustering classification model to provide a classification of the power materials, specifically including: Multiple target features are acquired and fused into a clustering feature vector; The manifold clustering distance is obtained by using a power material clustering classification model and a manifold clustering algorithm. Based on the manifold clustering distance, the power materials to be tested are clustered and classified, and the corresponding test classification of the power materials to be tested is given.

3. The method for generating power material inspection tasks according to claim 2, characterized in that, Based on manifold clustering distance, the electrical materials to be inspected are clustered and classified, and the corresponding inspection classification of the electrical materials to be inspected is given, specifically including: Clustering algorithm is selected based on manifold clustering distance; The number of clusters is determined by combining domain knowledge and effectiveness metrics; Based on the selected clustering algorithm, the power materials to be detected are merged into several clusters using manifold clustering distance as a metric. Analyze the feature distribution of each cluster to obtain the risk characteristics of each cluster; Based on the risk characteristics of each cluster, risk labels are given. By combining the risk labels and material types, the corresponding detection classification of the power materials to be tested is given.

4. The method for generating power material inspection tasks according to claim 1, characterized in that, The construction of the detection requirement model specifically includes: Obtain historical inspection reports and inspection requirement rule sets for various types of power equipment; The historical inspection reports of various types of power materials are analyzed using a semantic tensor extraction model for power materials, and target information is extracted to construct a standardized information database. By combining a standardized information database and a set of testing requirements rules, a testing demand model is constructed.

5. The method for generating power material inspection tasks according to claim 4, characterized in that, The target information includes the type of material, the testing items, and the testing results data; By combining a standardized information database and a set of testing requirements rules, a testing requirements model is constructed, which specifically includes: Data on material types, testing items, and testing results are obtained from a standardized information database. The set of testing requirement rules is transformed into a rule field to form a structured dataset, which is then divided into a training set and a validation set. Obtain the initial detection requirement model, and based on the training set, provide the rule field gravity, historical data potential energy, and rule deviation penalty; By combining the gravitational force of the rule field, the potential energy of historical data, and the penalty for rule deviation, the detection requirement field strength for each detection item is obtained; Based on the field strength required for detection, a set of candidate detection items is provided; The parameters of the initial detection requirement model are optimized through training, and then validated and adjusted using a validation set to obtain the detection requirement model.

6. The method for generating power material inspection tasks according to claim 1, characterized in that, The construction of the detection method library specifically includes: The testing method set includes a variety of testing methods and the corresponding testing items for each testing method; The detection attribute data includes resource consumption parameters and performance index data. Resource consumption parameters include detection equipment occupancy, manpower requirements, and time costs. Performance index data includes detection accuracy, applicable scenarios, and historical anomaly correlation. Dynamic detection adaptation rules include detection priority strategy and detection alternative strategy.

7. The method for generating power material inspection tasks according to claim 6, characterized in that, The construction of a multi-objective optimization function specifically includes: Based on a standardized information database and a set of testing requirements rules, a set of quality risks is given, and the coverage of the candidate testing item set to the quality risks is determined. Based on the upper limit of detection resources, obtain the fit between the candidate detection item set and the available resources; By combining the time cost ceiling, the time efficiency of the candidate detection item set is obtained; By combining coverage, adaptability, and time efficiency, a multi-objective optimization function is constructed, and constraints are set, including mandatory inspection item constraints, resource upper limit constraints, and time upper limit constraints.

8. The method for generating power material inspection tasks according to claim 7, characterized in that, A dynamic fitness field matching algorithm is used to match a corresponding detection method for each detection item in the optimal detection item set, specifically including: For each detection item in the optimal detection item set, obtain the detection performance, matching degree with real-time resources, and task urgency of each detection method associated with the detection item; By combining the detection efficiency of each detection method, its matching degree with real-time resources, and the urgency of the task, the dynamic adaptability of the detection method is obtained. The dynamic fit of each detection method is compared, and the detection method with the highest dynamic fit is selected as the detection method that matches the current detection item, until the matching of all detection items in the optimal detection item set is completed.

9. A power material inspection task generation device, employing the power material inspection task generation method as described in any one of claims 1-8, characterized in that, include: The detection clustering and classification module is used to analyze the information of the power materials to be detected, provide multiple target features, process the target features through the power material clustering and classification model, and provide the detection classification of the power materials. The inspection item acquisition module is used to obtain a set of candidate inspection items by combining the inspection classification of power materials and the pre-built inspection requirement model. The detection item optimization module is used to provide the optimal set of detection items based on a pre-established detection method library, using a multi-objective optimization function and iterative optimization through a genetic algorithm. The detection method matching module combines real-time detection resources and time costs, and employs a dynamic fit field matching algorithm to match a corresponding detection method for each detection item in the optimal detection item set. Specifically, it includes: generating a finite number of effective combined detection methods based on the detection methods in the detection method library; obtaining the dynamic fit of each combined detection method by combining its detection performance, matching degree with real-time resources, task urgency, tunneling probability, and combined gain; comparing the dynamic fit of each combined detection method, and selecting the combined detection method with the highest dynamic fit as the combined detection method matched with the current detection item. The dynamic fit of the combined detection method is specifically expressed as follows: ； Among them, D j Eff(M) represents the dynamic fitness of the j-th combined detection method, where m is the total number of combined detection methods. j ) is a combined detection method M j The detection efficiency of Res(M) j (Z) is a combined detection method M j The matching degree with real-time resource Z, Urgent(T) is the task urgency, T represents the remaining time of the task, and P is the urgency of the task. tunnel It is the tunneling probability. It is the combined gain, and h is the resource adaptation constant. For resource equivalent quality, V barrier Where D represents the resource barrier and D represents the distance to resource substitution. These are the weighting coefficients. The difference in detection accuracy between the combined detection method and the detection method being replaced. T represents the resource consumption cost ratio of the combined detection method to the detection method being replaced. win Let n be the time window pressure, and n be the number of detection methods that make up the combined detection method. k Let be the confidence level of the k-th detection method. OP represents the order penalty factor, and OP represents the execution order penalty; generate a list of detection tasks.

Images