Unplanned operation risk detection method, device and equipment and storage medium

Abstract

The invention discloses an unplanned operation risk detection method and device, equipment and a storage medium, and the method comprises the steps: carrying out the structural processing of obtained original power operation and maintenance data, and obtaining the power operation and maintenance data; outputting an industry semantic vector set based on text content and visual content in the power operation and maintenance data and a pre-trained power industry semantic prior model; the electric power industry semantic priori model is generated based on training of electric power operation and maintenance data marked with industry semantics; inputting the industry semantic vector set into an unplanned operation risk detection model, and outputting an unplanned operation risk detection result corresponding to the power operation and maintenance data; the unplanned operation risk detection model is obtained based on combined training of a conditional generation diffusion model and a multi-view comparative learning network. By utilizing the method, the accuracy and timeliness of unplanned operation detection in the electric power operation and maintenance process are improved, and potential safety hazards and conflicts are effectively identified and prevented.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of power grid technology, and in particular to a method, device, equipment, and storage medium for detecting risks of unplanned operations. Background Art

[0002] In the field of power operations and maintenance, as the scale and complexity of power grids increase, ensuring the safe and efficient operation of infrastructure faces challenges, particularly in the detection and identification of unplanned operations. Power operations and maintenance involve a variety of data types, including text, images, video surveillance, and sensor data. These data are widely distributed in time and space and come in a variety of formats, increasing the complexity of data integration and utilization. Furthermore, operations and maintenance activities involve the collaboration of multiple disciplines, and interdepartmental planning conflicts often arise due to poor coordination. Therefore, traditional planning management systems are prone to inconsistencies between operations and records during actual operations and maintenance, resulting in unplanned operations.

[0003] Existing technologies primarily rely on a single data dimension and simple anomaly detection models, such as support vector machines or deep learning networks, for independent anomaly identification. They are insufficient for collaborative analysis and deep integration of multimodal data. Existing methods are often based on pre-set rules and limited data splicing, making them insensitive to emerging unplanned operation risks. Summary of the Invention

[0004] The embodiments of the present invention provide a method, apparatus, device and storage medium for detecting the risks of unplanned operations, which improve the accuracy and timeliness of unplanned operation detection during power operation and maintenance, and effectively identify and prevent potential safety hazards and conflicts.

[0005] In a first aspect, an embodiment of the present invention provides a method for detecting risks of unplanned operations, which is applied to a computer device, wherein the computer device is communicatively connected to an electronic scale, and includes:

[0006] Perform structured processing on the acquired raw power operation and maintenance data to obtain power operation and maintenance data;

[0007] Outputting an industry semantic vector set based on the textual content and visual content in the power operation and maintenance data and a pre-trained power industry semantic prior model; the power industry semantic prior model is generated by training the power operation and maintenance data labeled with industry semantics;

[0008] The industry semantic vector set is input into the unplanned operation risk detection model, and the unplanned operation risk detection result corresponding to the power operation and maintenance data is output; the unplanned operation risk detection model is obtained by combining training based on the conditional generative diffusion model and the multi-view contrastive learning network.

[0009] In a second aspect, an embodiment of the present invention further provides a device for detecting risks of unplanned operations, the device comprising:

[0010] The data processing module is used to perform structured processing on the acquired raw power operation and maintenance data to obtain power operation and maintenance data;

[0011] A vector output module is configured to output an industry semantic vector set based on the textual content and visual content in the power operation and maintenance data and a pre-trained power industry semantic prior model; the power industry semantic prior model is generated by training the power operation and maintenance data labeled with industry semantics;

[0012] The detection result output module is used to input the industry semantic vector set into the unplanned operation risk detection model and output the unplanned operation risk detection result corresponding to the power operation and maintenance data; the unplanned operation risk detection model is obtained based on the combined training of the conditional generative diffusion model and the multi-view contrastive learning network.

[0013] In a third aspect, an embodiment of the present disclosure further provides an electronic device, comprising:

[0014] one or more processors;

[0015] a storage device for storing one or more programs,

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the unplanned operation risk detection method provided by the embodiment of the present disclosure.

[0017] In a fourth aspect, an embodiment of the present disclosure further provides a storage medium comprising computer-executable instructions, which, when executed by a computer processor, are used to implement the unplanned operation risk detection method provided by an embodiment of the present disclosure.

[0018] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the unplanned operation risk detection method provided by the embodiment of the first aspect above.

[0019] The present invention discloses a method, apparatus, device, and storage medium for detecting the risk of unplanned operations. By performing structured processing on raw power operation and maintenance data, multi-source data can be fully utilized, improving the comprehensiveness of the data. A pre-trained power industry semantic prior model is used to extract textual and visual content from power operation and maintenance data, comprehensively capturing and understanding diverse operation and maintenance scenarios and hidden dangers, and improving the accuracy of anomaly identification. Analysis using an unplanned operation risk detection model based on a generative diffusion model and multi-perspective comparative learning improves the accuracy of detecting unplanned operation events. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the originals and elements are not necessarily drawn to scale.

[0021] Figure 1 A flowchart of a method for detecting risks of unplanned operations provided by an embodiment of the present disclosure;

[0022] Figure 2 This is a flow chart of a method for detecting risks of unplanned operations provided in the second embodiment of the present disclosure;

[0023] Figure 3 This is a flow chart of a method for training an unplanned operation risk detection model provided in the second embodiment of the present disclosure;

[0024] Figure 4 This is a flow chart of a method for training a basic model of an unplanned operation risk detection model provided in the second embodiment of the present disclosure;

[0025] Figure 5 A schematic structural diagram of an unplanned operation risk detection device provided by an embodiment of the present disclosure;

[0026] Figure 6 A schematic structural diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0027] Figure 1 A flowchart of unplanned operation risk detection is provided in an embodiment of the present disclosure. The embodiment of the present disclosure is applicable to providing a solution to the problem that power operation and maintenance is multi-dimensional and has different data formats, and the single data processing model is too simple. The method can be executed by an unplanned operation risk detection device, which can be implemented in the form of software and / or hardware.

[0028] like Figure 1 As shown, an embodiment of the present disclosure provides a method for detecting risks of unplanned operations, which may specifically include the following steps:

[0029] S101: Structural processing is performed on the acquired original power operation and maintenance data to obtain power operation and maintenance data.

[0030] In this embodiment, raw power operation and maintenance data may be raw data generated during the power operation and maintenance process. This raw data may include: professional plan texts and safety supervision work plan texts; operation process records and image files extracted from on-site inspection applications (Application Software, APP); video frames and temperature distribution maps collected from camera monitoring logs and infrared thermal imaging files; and coordinate information and equipment identification of each substation or transmission line extracted from a geographic information system. The power operation and maintenance data may be data that has undergone data cleaning, spatiotemporal alignment, data aggregation, or special data tagging.

[0031] Specifically, raw power operation and maintenance data can be obtained from a power operation and maintenance management system through a unified data acquisition interface or collected through an automatic monitoring system. The raw power operation and maintenance data is then processed using at least one of the following methods, such as data cleaning, spatiotemporal alignment, data aggregation, or tagging of specific data. The processed data is then used as power operation and maintenance data. This power operation and maintenance data may include the following elements: professional plan text, safety supervision operation text, data related to on-site inspection apps and camera monitoring, infrared thermal imaging data, and geographic location information and operation time periods.

[0032] S102: Based on the text content and visual content in the power operation and maintenance data and the pre-trained power industry semantic prior model, output an industry semantic vector set.

[0033] Among them, the semantic prior model of the power industry is generated by training based on power operation and maintenance data labeled with industry semantics.

[0034] In this embodiment, the power operation and maintenance data includes text content composed of text data, such as professional plan texts, safety supervision operation texts, etc. The power operation and maintenance data also includes visual content composed of video data or infrared imaging data, such as on-site inspection APP and camera monitoring related data, infrared thermal imaging data, etc. The power industry semantic prior model can be a pre-trained model that integrates domain knowledge (such as equipment topology, failure mode, safety specifications, etc.), which provides a priori constraints for anomaly detection by structured representation of industry semantic relationships. In this application, the power industry semantic prior model is generated by training a training set composed of power operation and maintenance data labeled with industry semantics. The industry semantic vector set can be a unified semantic vector formed by text content and visual content, and integrates the geographic location code and operation time label of the substation and line. The text content includes fault category index, operation safety level index, equipment information index, timestamp, geographic location code, and the visual content includes the number of operators, tool type and infrared hotspot scene information, which are formed after cross-modal mapping.

[0035] Specifically, the system first extracts textual and visual content from power operation and maintenance data. Textual content can be extracted through entity recognition based on a large power-specific language model. Another example is a speech-to-text system enhanced with semantically enhanced power terminology, which converts inspectors' spoken words into text and automatically completes specialized terminology by linking to a device library. Another example is document structured parsing. Visual content can be extracted through infrared thermal imaging analysis or video dynamic behavior analysis. The extracted textual content and visual content from the power operation and maintenance data are then combined with features from a pre-trained power industry semantic prior model, resulting in the output of an industry semantic vector set.

[0036] S103: Input the industry semantic vector set into the unplanned operation risk detection model, and output the unplanned operation risk detection result corresponding to the power operation and maintenance data.

[0037] In this embodiment, the unplanned operation risk detection model is trained based on a combination of a conditional generative diffusion model and a multi-view contrastive learning network. The training dataset can be a set of industry semantic vectors labeled with unplanned operation risks. The unplanned operation risk detection model includes a conditional generative diffusion network, a multi-view contrastive learning network, and a risk assessment network. The unplanned operation risk detection results can include risk scores for all records in the input raw power operation and maintenance data, as well as the causes and specific fields of possible unplanned operation risk anomalies. The specific fields represent information such as the time, equipment, or location of occurrence.

[0038] Specifically, the industry semantic vector set is input into the conditional generative diffusion network and the multi-perspective comparative learning network in the unplanned operation risk detection model respectively, and the potential risk and difference analysis results are output. Then, the output potential risk and difference analysis results are input into the risk assessment network. The risk assessment website performs a combined analysis of the above two and finally outputs the unplanned operation risk detection results.

[0039] The technical solution of the embodiments of the present invention, through structured processing of raw power operation and maintenance data, can fully utilize multi-source data and improve data comprehensiveness. A pre-trained power industry semantic prior model is used to extract textual and visual content from power operation and maintenance data, comprehensively capturing and understanding diverse operation and maintenance scenarios and hidden dangers, and improving the accuracy of anomaly identification. Analysis using an unplanned operation risk detection model based on a generative diffusion model and multi-view comparative learning improves the accuracy of detecting unplanned operation events.

[0040] Example 2

[0041] Figure 2The flowchart of a method for detecting risk of unplanned operations provided in the second embodiment of the present invention is further optimized and expanded based on the above embodiment, and can be combined with various optional technical solutions in the above embodiment. Figure 2 As shown, the second embodiment provides a method for detecting risks of unplanned operations, which specifically includes the following steps:

[0042] S201: Retain data whose evaluation results are greater than a first threshold value in the original power operation and maintenance data through a credibility evaluation function, and convert the retained data into a predefined data structure to obtain standard format data.

[0043] The credibility assessment function is used to quantify the reliability of the data, outputting a confidence score in the interval [0, 1]. The first threshold can be a pre-set value between 0 and 1, which can be set based on actual circumstances and is not specifically limited in this embodiment. The predefined data structure can include text, images, and corresponding timestamps. The standard format data can be the original power operation and maintenance data after the data structure is converted.

[0044] Specifically, the data in the original power operation and maintenance data is evaluated using a credibility evaluation function to obtain an evaluation result. The evaluation result is then compared with a preset first threshold. Data above the first threshold is retained, while data below the first threshold is deleted. The retained data is then uniformly encoded into a predefined data structure to obtain standard format data.

[0045] This method can be used to remove records with too many missing fields or obviously abnormal information, thereby enhancing the usability of the data.

[0046] For example, the original power operation and maintenance data is cleaned and formatted, records with too many missing fields or obvious abnormal information are removed, and the remaining data is uniformly encoded into a standard structure:

[0047] D c ={x i ∣x i ∈Ω,β(x i )≥θ}

[0048] where x i represents all the information of a candidate data record (including text, image and corresponding timestamp), β(x i ) represents the score of the record in the completeness and credibility evaluation, and θ is the threshold for data cleaning.

[0049] Convert all retained data into a unified data table structure:

[0050] D c ={r1,r2,…,r n}

[0051] where r i Represents a record after preliminary cleaning and structuring, including text fields, image fields, and basic metadata.

[0052] S202: Perform spatiotemporal alignment on the text data in the standard format data according to the device identifier and the operation time period of the standard format data.

[0053] In this embodiment, the data record in the standard format data includes the device identification of the device used for the recording and the operation time period of the recording operation, wherein the operation time period includes a start time and an end time.

[0054] Specifically, the text data of different contents in the standard format data are temporally and spatially aligned according to the equipment identification, the start time and the end time of the operation time period, wherein the text data may include the above-mentioned professional plan text record and the safety supervision operation plan text.

[0055] For example, for the reserved data table D c , using the equipment identifiers in the professional plan text and the safety supervision operation plan text, the start time and the end time are aligned.

[0056] The professional plan text record is:

[0057]

[0058] The text record of the safety supervision operation plan is:

[0059]

[0060] e i With e j Respectively represent the equipment identification e in the professional plan of Article i and the safety supervision plan of Article j i With e j , and Respectively represent the starting time of the i-th professional plan and the j-th safety supervision plan, and They represent the end time of the i-th professional plan and the j-th safety supervision plan respectively.

[0061] Based on this embodiment, it is also possible to determine whether there is overlap by using a time intersection function Γ:

[0062]

[0063] If and only if

[0064] e i =e j and

[0065] When the i-th professional plan text and the j-th safety supervision operation plan text are determined to overlap in equipment and time period, the mapping relationship can be stored.

[0066] S203 , performing an aggregation operation on the time-space aligned standard format data according to the geographic location information and the operation time period of the standard format data, and using the aggregated standard format data as power operation and maintenance data.

[0067] In this embodiment, the data record in the standard format data includes the geographical location information of the record.

[0068] Specifically, an aggregation operation is performed on the standard format data after time-space alignment according to the Euclidean distance between the geographical location and the operation time, and the standard format data after the aggregation operation is used as the power operation and maintenance data.

[0069] For example, the power operation aggregation function Ψ is as follows:

[0070] Ψ(r i ,r j )=δ(g i ,g j )×η(t i ,t j )

[0071] in:

[0072] δ(g i ,g j )=exp(-α·||g i -g j ||)

[0073]

[0074] ||g i -g j || represents the geographical location of the i-th professional plan Geographic location with respect to the Article j safety oversight plan The Euclidean distance between i ,g j ) represents the distance difference function, η(t i ,t j ) represents the time difference function, α and β are the attenuation factors determined in the scene. i ,r j ) is greater than the preset threshold ρ, then it is determined that the i-th professional plan text and the j-th safety supervision operation plan text are in an aggregated state in terms of geographical location and time period.

[0075] By using this method, the data is merged into the same aggregation cluster to achieve the classification management of operation records in the same time and space area.

[0076] Based on the above embodiment, when the same equipment appears in both the professional plan text and the safety supervision operation plan text, and the number of records that satisfy the time intersection function Γ>0 in the time period exceeds a threshold, a potential multi-professional parallel conflict is determined. For on-site inspection app records, if camera monitoring logs, camera monitoring video data, and text records still match the same cluster after aggregation, this information is marked as a potential anomaly or a key concern in the corresponding record.

[0077] S204: Obtain a predefined electric power industry dictionary.

[0078] In this embodiment, the power industry dictionary may be a mapping table related to the power industry, wherein the mapping relationships included in the mapping table may include fault category, equipment name, operation measure, operation safety level, etc.

[0079] Specifically, a predefined electric power industry dictionary is obtained.

[0080] For example, let the power industry dictionary Λ contain four parts: fault category, equipment name, operation measures, and operation safety level:

[0081] Λ={Λ f ,Λ e ,Λ m ,Λ s}

[0082] where Λ f List of standard mappings representing fault categories, Λ e A list of standard mappings representing device names, Λ m A list of standard mappings representing operational measures, Λ s A list of standard mappings representing job security levels.

[0083] S205 . Map the text content in the power operation and maintenance data uniformly according to the power industry dictionary to obtain standardized text.

[0084] Specifically, within the acquired power operation and maintenance data, the professional plan text and the safety supervision work plan text are mapped uniformly using the power industry dictionary Λ to map fault categories, equipment names, work measures, and work safety levels. The corresponding spatiotemporal codes are then incorporated to generate standardized text. The spatiotemporal codes can be obtained by encoding the geographic location information and the work time period.

[0085] Suppose any text record is Contains the original key field (f k ,e k ,m k ,s k ), corresponding to fault category, equipment name, operation measures and operation safety level respectively, then the standardized conversion function is:

[0086]

[0087] where φ f (·),φ e (·),φ m (·),φ s (·) respectively represent the f ,Λ e ,Λ m ,Λ s After the mapping is completed, a standardized fault category index, equipment name index, operation measure index and safety level index are obtained in each text record.

[0088] Based on the result and the geographic location code G(g lat ,g lon ) and operation time period T(t start ,t end ) Combined and unified in recording Stored as

[0089]

[0090] S206 : Extracting visual content from the power operation and maintenance data as visual features according to a predefined visual feature extraction function.

[0091] In this embodiment, the visual feature extraction function can be a mathematical tool for converting image / video data into a structured feature vector. The visual feature can be a feature encoded by the visual content in the power operation and maintenance data.

[0092] Specifically, the visual content in the power operation and maintenance data is extracted according to a predefined visual feature extraction function as visual features, and the visual features are stored in a structure corresponding to the text standardization record to achieve a one-to-one correspondence between image information and text information.

[0093] For example, the visual feature extraction function Ψ is defined as v Quantify the number of workers, types of tools, and infrared abnormal hot spots in the monitoring image or infrared image:

[0094]

[0095] in Represents the visual data entry of the video in the power operation and maintenance data, Θ v Represents a set of visual detection parameters specifically used to identify the scene, The output visual feature vector includes the number of workers counted, tool type index, and infrared hotspot value distribution. The data is stored in a structure corresponding to the text standardization record, so as to realize a one-to-one correspondence between the image information and the text information.

[0096] S207. Using the power industry semantic prior model, the standardized text and visual features are spliced into the same vector space and encoded to obtain an industry semantic vector set.

[0097] Specifically, the text and visual features are mapped to a unified dimensional space, the electric power knowledge graph is introduced as a loss function constraint, the features mapped to the unified dimensional space are gated and fused, and the industry semantic vector set is output.

[0098] Using this method, the industry semantic prior on the text side is aligned with the real-time detection information on the visual side in the vector space.

[0099] For example, for the same record r k All information in the text domain and image domain is fused across modal vectors, denoted as z k .

[0100] Let the text normalized output be The visual feature output is The fusion function is:

[0101]

[0102] where Ω cm (·) represents a cross-modal vectorized mapping operator, which is used to concatenate textual information and visual information into the same vector space and perform encoding operations.

[0103] make In the text domain has dimension d txt , In the visual domain, it has dimension d img . Connect the two into an initial fusion vector (Dimension d txt +d img ), through the nonlinear transformation F nl , the learnable parameter W is used to obtain the final cross-modal vector z k :

[0104]

[0105] W represents the trainable weight matrix for cross-modal mapping, Fnl Represents the nonlinear activation function introduced in the fusion process. After completing the above cross-modal vectorization, the z of all records is summarized. k :

[0106]

[0107] Each element z in k It also includes a unified semantic vector formed by standardized features on the text side and visual features, and integrates the geographic location codes and operation time labels of substations and lines. The output is a multimodal spatiotemporal fusion industry semantic vector set, which includes fault category index, operation safety level index, equipment information index, timestamp, and geographic location code in the text domain, and the number of operators, tool type, and infrared hotspot scene information in the visual domain.

[0108] S208. Obtain a pre-trained unplanned operation risk detection model; wherein the unplanned operation risk detection model includes: a conditional generative diffusion network, a multi-view contrastive learning network and a risk assessment network.

[0109] This step is used to obtain an unplanned operation risk detection model. The model includes a conditional generative diffusion network, a multi-view contrastive learning network, a risk assessment network, and an input layer and an output layer. The input layer is used to input data, and the output layer is used to output results.

[0110] S209: Input the industry semantic vector set into the conditional generative diffusion network and output the potential risk vector set.

[0111] Specifically, the temporal information and spatial information of the industry semantic vector set and the predefined industry prior information are input into the conditional generation diffusion network as conditional information, and the potential risk vector set is output.

[0112] S210: Input the industry semantic vector set into the multi-view comparative learning network and output the difference analysis results.

[0113] This step is used to input the industry semantic vector set into the multi-view contrastive learning network and output the difference analysis results.

[0114] Among them, multi-perspective comparison includes at least one of the following: 1. Professional plan and safety supervision plan perspective in the text content: compare whether the records of professional plan and safety supervision plan are consistent within the same time period of the same station or the same equipment; 2. Text content and on-site image and video information perspective: if the text does not record power outage maintenance or major faults, but the on-site image or video data shows that there are personnel operations or equipment abnormalities, it is considered that there is an obvious conflict; 3. Cross-station comparison perspective: for the same type of major faults or important maintenance plans, check their registration consistency between different stations; 4. Cross-professional comparison perspective: compare whether there are conflicts or omissions at the time or equipment level in the registration of dispatching, substation, transmission and other professions.

[0115] S211. Input the potential risk vector set and the difference analysis results into the risk assessment network, and output the unplanned operation risk detection results.

[0116] In this embodiment, the risk assessment network is a functional module that combines the outputs of two different neural networks to output a detection result. The core concept of this functional module is: if the two different neural networks agree on the same conflict, they will be amplified; if there is a clear disagreement, the traceability mechanism will be triggered and the two networks will be smoothly merged.

[0117] Specifically, the potential risk vector set and the difference analysis results are input into the risk assessment network. For the same record in the power operation and maintenance data, the difference strength between the results output by two different neural networks is determined. Based on the difference strength, different operation steps are determined and the unplanned operation risk detection results are output. The operation steps include amplifying the difference strength or triggering the traceability mechanism and smoothing the fusion.

[0118] Furthermore, based on the above-mentioned embodiment of the invention, the potential risk vector set and the difference analysis result are input into the risk assessment network, and the method of outputting the unplanned operation risk detection result also includes:

[0119] S2110. For the same record in the power operation and maintenance data, determine, according to a quantization function, a first difference strength corresponding to the potential risk vector set and a second difference strength corresponding to the output difference analysis result;

[0120] S2111: If the difference between the first difference strength and the second difference strength is less than a second threshold, increase the first difference strength and the second difference strength; if the difference between the first difference strength and the second difference strength is greater than or equal to the second threshold, mark the record as pending verification;

[0121] S2112. Determine a comprehensive risk score for each record based on the first difference strength and the second difference strength and use it as the unplanned operation risk detection result.

[0122] In this embodiment, the first difference strength can reflect the degree of dispersion of the potential risk vector and is obtained by calculating the volatility of the risk probability value. The second difference strength reflects the degree of deviation of the result and is obtained by comparing the mean of the absolute differences of the probability values item by item. The second threshold can be a pre-set threshold value, which is set according to the actual situation.

[0123] Specifically, for the same record in the power operation and maintenance data, the first difference intensity corresponding to the potential risk vector set and the second difference intensity corresponding to the output difference analysis result are determined according to the quantization function. If the difference between the first difference intensity and the second difference intensity is less than the second threshold, it means that the risk feature is fuzzy or the disagreement between AI and experts is small. At this time, the difference intensity between the two is dynamically enhanced through the time attenuation factor to avoid missed detection. If the difference is greater than or equal to the second threshold, it indicates that there is a significant contradiction between the AI and expert judgments, and it is directly marked as a pending verification state. Finally, based on the first difference intensity and the second difference intensity, combined with predefined prior operators and difference weights and other information, a comprehensive risk score for each record is finally obtained as the unplanned operation risk detection result. Among them, the prior operator can include: equipment type, fault level and safety level.

[0124] For example, the heterogeneous conflict strength output by the potential risk vector set on the kth record is the first difference strength The comprehensive difference intensity of the output difference analysis result of the multi-view contrast learning network on the kth record is the second difference intensity Differentiation for:

[0125]

[0126] Ψ diff (·) represents the function of quantizing the first difference intensity, Ψ cont (·) represents the function of quantizing the second difference intensity.

[0127] When two different network modules are integrated, and If both are higher than their respective thresholds and the identified spatial or professional conflicts point to the same area (or the same time period, the same equipment), then the record can be considered to have higher credibility and priority in terms of unplanned operation risk;

[0128] If the two identify a clear conflict, the geographical location, operation time period, and visual information of the record must be traced back to check for registration errors or noise interference. If two different network modules agree on the same conflict, they will be amplified; if there is a clear disagreement, the traceability mechanism will be triggered and smooth integration will be achieved:

[0129]

[0130] in, represents a collaborative amplification operation, Indicates conflict smoothing operation. If two different network modules give conflicting judgments on the same record, the maximum value of the two will be reduced first, and then constraint fusion will be performed. It is the traceability penalty coefficient. When a conflict occurs and it is difficult to determine after manual or automatic rule verification, the original registration information and multimodal data are further traced back and the record is temporarily listed as pending verification. Further combining the prior operator feedback, difference weight and other information introduced in the introduction, the comprehensive risk score of each record is finally obtained. This score is the comprehensive risk value of the unplanned operation, recorded as

[0131]

[0132] Represents the text standardization information; Λ(·) is the weight function extracted from the safety level, work measures, etc. For example, if the safety level is low or the fault level is high, Λ(·) will increase. Ω dual (·) is the output conflict measure of dual-different network modules fusion, represents the industry prior information introduced in step S3, and Γ(·) is the prior correction function used to further increase (or decrease) the risk value of high-priority equipment and faults.

[0133] Based on the above embodiment, a method for detecting risks of unplanned operations further includes: a training process of an unplanned operation risk detection model;

[0134] like Figure 3 As shown in Figure 2, the training process of the unplanned operation risk detection model includes:

[0135] S301. Obtain an industry semantic vector set labeled with unplanned operation risks as a training dataset.

[0136] In this embodiment, an industry semantic vector set marked with unplanned operation risks determined through historical power operation and maintenance data is obtained as a training data set.

[0137] S302: Divide the training data set into a training set and a test set according to a preset data set segmentation ratio, and initialize and construct an initial model of the unplanned operation risk detection model.

[0138] S303: Train the initial model of the unplanned operation risk detection model according to the training set to obtain a basic model of the unplanned operation risk detection model.

[0139] In this embodiment, the preset data set segmentation ratio may be a pre-set segmentation ratio, which is set according to actual conditions. The unplanned operation risk detection model base model may be a base model that performs multiple rounds of iterations on the training set to gradually optimize the model parameters.

[0140] Specifically, the training data set is divided into a training set and a test set according to a preset data set segmentation ratio, and the initial model of the unplanned operation risk detection model is initialized and constructed. Through gradient descent or other optimization algorithms, a suitable optimization function is selected, the model parameters of the initial neural network are adjusted, and multiple rounds of iterations are performed on the training set to gradually optimize the model parameters and obtain the basic model of the unplanned operation risk detection model.

[0141] S304. Use the test set to perform a performance test on the basic model of the unplanned operation risk detection model; if the test result meets the preset model test pass conditions, the basic model of the unplanned operation risk detection model will be used as the final unplanned operation risk detection model, otherwise the initial model of the unplanned operation risk detection model will be retrained according to the training set.

[0142] Specifically, the test set is used to evaluate the performance of the basic model of the unplanned operation risk detection model, such as prediction accuracy, generalization ability and computational efficiency. The preset test pass conditions include but are not limited to the prediction error threshold, convergence speed and resource consumption. If the test results meet the pass conditions, the basic model of the unplanned operation risk detection model will be used as the final model; otherwise, the initial model will be retrained until the conditions are met.

[0143] On the basis of the above embodiment, the initial model of the unplanned operation risk detection model is trained according to the training set to obtain the basic model of the unplanned operation risk detection model, such as Figure 4 As shown, including:

[0144] S3031. Use the temporal information and spatial information of the historical industry semantic vector set and the predefined industry prior information as conditional information, use the difference function as the loss function, and optimize the model parameters of the conditional generative diffusion network in the initial model of the unplanned operation risk detection model with the goal of minimizing the difference function to obtain the basic conditional generative diffusion network.

[0145] An example of a training situation is as follows:

[0146] In the conditional generative diffusion network, the forward process of noise diffusion and the generative process of reverse sampling on the noise distribution are introduced. The conditional information in this application is the temporal information and spatial information of the historical industry semantic vector set.

[0147] Let x0 represent the multimodal vector representation in the original noise-free state, x trepresents the noisy vector after the t-th iteration, t∈{1,2,…,T}, T is the number of diffusion steps.

[0148] The forward diffusion process is:

[0149]

[0150] where β t represents the noise coefficient added in step t, and I is the unit matrix of the same dimension.

[0151] The reverse sampling process is approximated by introducing learnable parameters θ and conditional information c:

[0152]

[0153] In the reverse sampling process, let the geographical location contained in the historical industry semantic vector set be identified as g k , time code is t k , g k and t k As a space-time constraint operator, μ θ and Σ θ Make corrections, space-time constraint correction function:

[0154] F st (x t ,g k ,t k )=ξ(Δ location (g k ,g ref )+Δ time (t k ,t ref ))

[0155] where g ref and t ref Indicates the geographical location and time period information that need to be compared currently, Δ location (·) and Δ time (·) measure g respectively k With g ref ,t k With t ref ξ is the magnification factor.

[0156] If Δ location +Δ time If the value of is small, it means that there is a suspected parallel conflict at the same location or in the same time period, and it is necessary to perform reverse sampling on x. t The sampling mean of is disturbed to increase the visibility of the conflicting feature in the generation process:

[0157]

[0158] λ st is the weight of the spatiotemporal constraint operator. This approach can enhance multi-disciplinary parallel conflict information and highlight cross-regional registration anomalies.

[0159] Then, the equipment type, fault level and safety level are injected into the historical industry semantic vector set and used as a priori operators to participate in controllable sampling during back diffusion.

[0160] Let p k Represents the prior vector formed by the device type, fault level and safety level. This application proposes a controllable sampling correction function:

[0161] F prior (x t ,p k )=α dev σ dev (p k )+α flt σ flt (p k )+α safe σ safe (p k )

[0162] Among them, σ dev , represents the device type, σ flt, represents the fault level, σ safe Indicates the security level to perform numerical decoding, α dev ,α flt ,α safe is the adjustment coefficient of the controllable sampling parameter.

[0163] If σ flt (p k ) reflects that the fault level is extremely high, or σ safe (p k ) points out that if the security level is low, the discreteness caused by the reverse noise should be reduced during controllable sampling to avoid disordered results that completely violate industry common sense.

[0164] And further in Σ θ Add the following constraints:

[0165]

[0166] λ prior This is used to control the attenuation of the prior constraint in the variance term. This allows for guided corrections to the noise sampling and reconstruction process during the back-diffusion phase, preventing results from significantly deviating from common operational knowledge.

[0167] The historical industry semantic vector set zk The visual features contained in The number of workers, tool types and infrared abnormal hotspot distribution in the visual sampling correction function

[0168]

[0169] in Indicates the quantity characteristics corresponding to the operators and tools, represents the distribution characteristics of infrared abnormal hot spots, γ a with γ b If the visual information shows that there are many people or serious equipment temperature anomalies, an additional offset is added to the mean term of the back diffusion:

[0170]

[0171] is the mean after including the time and space constraints, λ vis is the weight of visual feature amplification. θ Add additional offsets to increase the proportion of visual anomalies in the generated results.

[0172] The difference function is used as the loss function, and the difference function is:

[0173]

[0174] Among them, ||·||1 is the L1 norm, ω k The difference weights are set based on the conflict between the device dimension, the spatiotemporal dimension, and the visual dimension. The model parameters of the conditional generation diffusion network in the initial model of the unplanned operation risk detection model are optimized with the goal of minimizing the difference function.

[0175] S3032. Extract the multi-perspective feature vector from the historical industry semantic vector set through at least two independent feature encoders, use the multi-perspective contrast loss function as the loss function, and optimize the model parameters of the multi-perspective contrast learning network in the initial model of the unplanned operation risk detection model with the goal of minimizing the multi-perspective contrast loss function to obtain a basic multi-perspective contrast learning network.

[0176] S3033. Combine the basic condition generation diffusion network and the basic multi-view comparative learning network to generate a basic model of the unplanned operation risk detection model.

[0177] An example of a training situation is as follows:

[0178] The multi-perspectives of this application include professional-security comparison, text-image comparison, cross-site comparison, and cross-professional comparison. Under each perspective, a difference measurement function is defined to calculate the degree of difference between the record and other records under that perspective. Difference amplification operator D ν :

[0179]

[0180] Υ ν (·) represents the feature embedding function under the viewing angle ν, ||·|| represents the L2 norm, ω ν is the reference weight of the viewing angle ν, Φ ν (·,·) is the scene difference amplification factor, which is used to amplify conflicts that are easily overlooked but are actually very indicative from this perspective.

[0181] For example, from the perspective of professional plan vs. safety supervision plan, if the equipment ID is the same and the time period is highly overlapping but one party has no maintenance record, then Φ ν The value will be significantly increased, thereby amplifying the impact of the conflict on the final difference.

[0182] The multi-view contrast loss function is used to guide the network to continuously amplify significant conflicts and weaken normal records during training or optimization. The heterogeneous feature vectors of a batch of training samples are denoted as

[0183] For any two of them and Pairing, defining cross-scene multi-view contrast loss for:

[0184]

[0185] Among them, D ν (·) is the perspective difference magnification operator defined in the previous step, l(·) represents the basic contrast error measurement function, It is a dynamic weight. Pro-Sec, Txt-Img, Station, and Discipline represent professional-security comparison, text-image comparison, cross-site comparison, and cross-professional comparison perspectives.

[0186] like and If the perspective ν is inconsistent but should be consistent, then Increase; if it should be different and is detected to be different, then By minimizing the multi-view contrast loss function, the model parameters of the multi-view contrast learning network in the initial model of the unplanned operation risk detection model are optimized to obtain the basic multi-view contrast learning network.

[0187] Example 3

[0188] Figure 5 The present invention also provides a schematic diagram of the structure of an unplanned operation risk detection device. Figure 5 As shown, the device includes: a data processing module 401, a vector output module 402 and a detection result output module 403.

[0189] The data processing module 401 is used to perform structured processing on the acquired original power operation and maintenance data to obtain power operation and maintenance data;

[0190] A vector output module 402 is configured to output an industry semantic vector set based on the textual content and visual content in the power operation and maintenance data and a pre-trained power industry semantic prior model; the power industry semantic prior model is generated by training the power operation and maintenance data annotated with industry semantics;

[0191] The detection result output module 403 is used to input the industry semantic vector set into the unplanned operation risk detection model and output the unplanned operation risk detection result corresponding to the power operation and maintenance data; the unplanned operation risk detection model is obtained based on the combined training of the conditional generative diffusion model and the multi-view contrastive learning network.

[0192] The technical solutions provided by the disclosed embodiments utilize structured processing of raw power operation and maintenance data, fully utilizing multi-source data and improving data comprehensiveness. A pre-trained power industry semantic prior model extracts textual and visual content from power operation and maintenance data, comprehensively capturing and understanding diverse operation and maintenance scenarios and hidden dangers, and improving the accuracy of anomaly identification. Analysis using an unplanned operation risk detection model based on a generative diffusion model and multi-perspective comparative learning improves the accuracy of detecting unplanned operation events.

[0193] Furthermore, the data processing module 401 may be used to:

[0194] Retain data whose evaluation results are greater than a first threshold value in the original power operation and maintenance data through a credibility evaluation function and convert the retained data into a predefined data structure to obtain standard format data;

[0195] Performing spatiotemporal alignment on the text data in the standard format data according to the device identification and the operation time period of the standard format data;

[0196] An aggregation operation is performed on the time-space aligned standard format data according to the geographic location information and the operation time period of the standard format data, and the standard format data after the aggregation operation is used as power operation and maintenance data.

[0197] Furthermore, the vector output module 402 may be used to:

[0198] Get predefined power industry dictionaries;

[0199] uniformly mapping the text content in the power operation and maintenance data according to the power industry dictionary to obtain standardized text;

[0200] extracting visual content from the power operation and maintenance data as visual features according to a predefined visual feature extraction function;

[0201] The standardized text and the visual features are spliced into the same vector space through the power industry semantic prior model and a dimension reduction or encoding operation is performed to obtain an industry semantic vector set.

[0202] Furthermore, the detection result output module 403 can be used to:

[0203] Obtaining the pre-trained unplanned operation risk detection model; wherein the unplanned operation risk detection model includes: a conditional generative diffusion network, a multi-view contrastive learning network and a risk assessment network;

[0204] Inputting the industry semantic vector set into the conditional generative diffusion network and outputting a potential risk vector set;

[0205] Inputting the industry semantic vector set into the multi-view contrast learning network and outputting a difference analysis result;

[0206] The potential risk vector set and the difference analysis result are input into the risk assessment network, and the unplanned operation risk detection result is output.

[0207] Furthermore, the detection result output module 403 can be used to:

[0208] For the same record in the power operation and maintenance data, determining, according to a quantization function, a first difference strength corresponding to the potential risk vector set and a second difference strength corresponding to the output difference analysis result;

[0209] If the difference between the first difference strength and the second difference strength is less than a second threshold, increase the first difference strength and the second difference strength; if the difference between the first difference strength and the second difference strength is greater than or equal to the second threshold, mark the record as pending verification;

[0210] A comprehensive risk score for each record is determined based on the first difference strength and the second difference strength and used as the unplanned operation risk detection result.

[0211] Furthermore, the device also includes: a training process of the unplanned operation risk detection model, including:

[0212] Obtain an industry semantic vector set that annotates the risks of unplanned operations as a training dataset;

[0213] Divide the training data set into a training set and a test set according to a preset data set segmentation ratio, and initialize and construct an initial model of the unplanned operation risk detection model;

[0214] The initial model of the unplanned operation risk detection model is trained according to the training set to obtain a basic model of the unplanned operation risk detection model;

[0215] The test set is used to perform a performance test on the basic model of the unplanned operation risk detection model; if the test result meets the preset model test pass condition, the basic model of the unplanned operation risk detection model is used as the final unplanned operation risk detection model, otherwise the initial model of the unplanned operation risk detection model is re-trained according to the training set.

[0216] Furthermore, the device further comprises:

[0217] Taking the temporal information and spatial information of the historical industry semantic vector set and predefined industry prior information as conditional information, and the difference function as the loss function, the model parameters of the conditional generative diffusion network in the initial model of the unplanned operation risk detection model are optimized with the goal of minimizing the difference function to obtain a basic conditional generative diffusion network;

[0218] Extracting a multi-view feature vector from the historical industry semantic vector set through at least two independent feature encoders, using a multi-view contrast loss function as a loss function, and optimizing the model parameters of the multi-view contrast learning network in the initial model of the unplanned operation risk detection model with the goal of minimizing the multi-view contrast loss function to obtain a basic multi-view contrast learning network;

[0219] The basic condition generation diffusion network and the basic multi-view contrast learning network are combined to generate the basic model of the unplanned operation risk detection model.

[0220] The above device can execute the methods provided by all the above embodiments of the present invention, and has the corresponding functional modules and beneficial effects of executing the above methods. For technical details not fully described in this embodiment, please refer to the methods provided by all the above embodiments of the present invention.

[0221] Example 4

[0222] Figure 6A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is provided. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0223] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0224] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0225] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any other suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the unplanned operation risk detection method.

[0226] In some embodiments, the unplanned operation risk detection method can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the unplanned operation risk detection method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute the unplanned operation risk detection method in any other suitable manner (e.g., via firmware).

[0227] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0228] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0229] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0230] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0231] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0232] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0233] The above specific implementation manner does not constitute a limitation on the protection scope of the present invention.

Claims

1. A method for detecting risks of unplanned operations, characterized in that: include: Perform structured processing on the acquired raw power operation and maintenance data to obtain power operation and maintenance data; Outputting an industry semantic vector set based on the textual content and visual content in the power operation and maintenance data and a pre-trained power industry semantic prior model; the power industry semantic prior model is generated by training the power operation and maintenance data labeled with industry semantics; Inputting the industry semantic vector set into an unplanned operation risk detection model, and outputting an unplanned operation risk detection result corresponding to the power operation and maintenance data; The unplanned operation risk detection model is obtained by combining training based on a conditional generative diffusion model and a multi-view contrastive learning network.

2. The method according to claim 1, characterized in that The structured processing of the acquired original power operation and maintenance data to obtain the power operation and maintenance data includes: Retaining data whose evaluation results are greater than a first threshold value in the original power operation and maintenance data through a credibility evaluation function and converting the retained data into a predefined data structure to obtain standard format data; Performing spatiotemporal alignment on the text data in the standard format data according to the device identification and the operation time period of the standard format data; An aggregation operation is performed on the time-space aligned standard format data according to the geographic location information and the operation time period of the standard format data, and the standard format data after the aggregation operation is used as power operation and maintenance data.

3. The method according to claim 1, characterized in that The outputting of an industry semantic vector set based on the text content and visual content in the power operation and maintenance data and a pre-trained power industry semantic prior model includes: Get predefined power industry dictionaries; uniformly mapping the text content in the power operation and maintenance data according to the power industry dictionary to obtain standardized text; extracting visual content from the power operation and maintenance data as visual features according to a predefined visual feature extraction function; The standardized text and the visual features are spliced into the same vector space through the power industry semantic prior model and an encoding operation is performed to obtain an industry semantic vector set.

4. The method according to claim 1, wherein The step of inputting the text-visual fusion feature into the unplanned operation risk detection model and outputting the unplanned operation risk detection result corresponding to the power operation and maintenance data includes: Obtaining the pre-trained unplanned operation risk detection model; wherein the unplanned operation risk detection model includes: a conditional generative diffusion network, a multi-view contrastive learning network and a risk assessment network; Inputting the industry semantic vector set into the conditional generative diffusion network and outputting a potential risk vector set; Inputting the industry semantic vector set into the multi-view contrast learning network and outputting a difference analysis result; The potential risk vector set and the difference analysis result are input into the risk assessment network, and the unplanned operation risk detection result is output.

5. The method according to claim 4, characterized in that The step of inputting the potential risk vector set and the difference analysis result into the risk assessment network and outputting the unplanned operation risk detection result includes: For the same record in the power operation and maintenance data, determining, according to a quantization function, a first difference strength corresponding to the potential risk vector set and a second difference strength corresponding to the output difference analysis result; If the difference between the first difference strength and the second difference strength is less than a second threshold, increase the first difference strength and the second difference strength; if the difference between the first difference strength and the second difference strength is greater than or equal to the second threshold, mark the record as pending verification; A comprehensive risk score for each record is determined based on the first difference strength and the second difference strength and used as the unplanned operation risk detection result.

6. The method according to claim 4, characterized in that The training process of the unplanned operation risk detection model includes: Obtain a historical industry semantic vector set annotated with unplanned operation risks as a training dataset; Divide the training data set into a training set and a test set according to a preset data set segmentation ratio, and initialize and construct an initial model of the unplanned operation risk detection model; The initial model of the unplanned operation risk detection model is trained according to the training set to obtain a basic model of the unplanned operation risk detection model; The test set is used to perform a performance test on the basic model of the unplanned operation risk detection model; if the test result meets the preset model test pass condition, the basic model of the unplanned operation risk detection model is used as the final unplanned operation risk detection model, otherwise the initial model of the unplanned operation risk detection model is re-trained according to the training set.

7. The method according to claim 4, characterized in that The initial model of the unplanned operation risk detection model is trained according to the training set to obtain a basic model of the unplanned operation risk detection model, including: Taking the temporal information and spatial information of the historical industry semantic vector set and predefined industry prior information as conditional information, and the difference function as the loss function, the model parameters of the conditional generative diffusion network in the initial model of the unplanned operation risk detection model are optimized with the goal of minimizing the difference function to obtain a basic conditional generative diffusion network; Extracting a multi-view feature vector from the historical industry semantic vector set through at least two independent feature encoders, using a multi-view contrast loss function as a loss function, and optimizing the model parameters of the multi-view contrast learning network in the initial model of the unplanned operation risk detection model with the goal of minimizing the multi-view contrast loss function to obtain a basic multi-view contrast learning network; The basic condition generation diffusion network and the basic multi-view contrast learning network are combined to generate the basic model of the unplanned operation risk detection model.

8. An unplanned operation risk detection device, characterized in that: include: The data processing module is used to perform structured processing on the acquired raw power operation and maintenance data to obtain power operation and maintenance data; A vector output module is configured to output an industry semantic vector set based on the textual content and visual content in the power operation and maintenance data and a pre-trained power industry semantic prior model; the power industry semantic prior model is generated by training the power operation and maintenance data labeled with industry semantics; A detection result output module, configured to input the industry semantic vector set into an unplanned operation risk detection model and output an unplanned operation risk detection result corresponding to the power operation and maintenance data; The unplanned operation risk detection model is obtained by combining training based on a conditional generative diffusion model and a multi-view contrastive learning network.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the unplanned operation risk detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the unplanned operation risk detection method according to any one of claims 1 to 7 when executed.

Images