A power plant data governance method, device, electronic equipment and program product

Abstract

The application discloses a power plant data governance method and device, electronic equipment and program product, and relates to the technical field of data governance. The method comprises the following steps: acquiring historical power plant data; performing keyword extraction, classifying the extracted keywords according to power plant types, and obtaining a first historical power plant data set; performing clustering processing on the first historical power plant data set, and obtaining a second historical power plant data set classified according to business types under each power plant type; expanding the second historical power plant data set to generate a historical training sample set; constructing a power plant data classification model based on the historical training sample set, wherein the initial network parameters of the model are determined by grey wolf optimization algorithm optimization; inputting real-time power plant data into the model to obtain a classification result and store the classification result. The technical scheme of the application can efficiently and accurately process multi-source heterogeneous power plant data, and realize unified data governance across power plants and across businesses.

Description

Technical Field

[0001] This application relates to the field of data governance technology, and in particular to a power plant data governance method, apparatus, electronic equipment, and program product. Background Technology

[0002] In the field of power plant data governance, the unified management and classification of massive heterogeneous data is a key issue. Traditional governance methods often employ manual standardization or simple differentiation strategies based on preliminary features, which are insufficient to address the complex and dynamic data governance needs of various types of power plants. Existing technical solutions still face significant shortcomings in practical applications: such as severe data silos, inconsistent classification standards, high human resource costs, and the inability to obtain deep data features, resulting in low classification accuracy and efficiency, and model training easily getting trapped in local optima, making it difficult to achieve unified management and efficient utilization of power plant data.

[0003] Therefore, there is an urgent need for a power plant data governance method that can efficiently and accurately process multi-source heterogeneous power plant data and achieve unified data governance across power plants and business operations. Summary of the Invention

[0004] This application provides a power plant data governance method, apparatus, electronic device, and program product.

[0005] In a first aspect, embodiments of this application provide a power plant data governance method, including: Obtain historical power plant data; Keyword extraction is performed on historical power plant data, and the historical power plant data is classified according to power plant type based on the extracted keywords, resulting in the first historical power plant dataset divided by power plant type. Clustering is performed on the first historical power plant dataset to obtain the second historical power plant dataset, which is divided by business type under each power plant type. The second historical power plant dataset was augmented to generate historical training sample sets for each business type. Based on historical training sample sets, a power plant data classification model is constructed using a deep learning model; the initial network parameters of the deep learning model are determined by optimization using the Grey Wolf optimization algorithm. Real-time power plant data is input into the power plant data classification model to obtain classification results, and the real-time power plant data is stored based on the classification results.

[0006] In some embodiments, historical power plant data includes at least one of audio data, image data, and character data.

[0007] In some embodiments, keyword extraction is performed on historical power plant data, including: Based on the word frequency-inverse text frequency index-discrete factor algorithm, feature words are extracted from historical power plant data, and the weight of each feature word is determined. The feature words are sorted according to their weights, and a preset number of feature words are selected as keywords for historical power plant data.

[0008] In some embodiments, the clustering process is based on the fuzzy C-means clustering algorithm and clusters according to an adaptive fuzziness parameter and a feature-weighted distance metric.

[0009] In some embodiments, data augmentation of the second historical power plant dataset includes: Based on the results of clustering, the samples in the second historical power plant dataset are divided into positive samples and negative samples; The positive samples are augmented based on the normal distribution rule to obtain augmented positive samples. Negative samples are augmented based on random distribution rules to obtain augmented negative samples. The augmented positive and negative samples are combined to form a historical training sample set.

[0010] In some embodiments, the deep learning model employs a long short-term memory network; The Grey Wolf Optimization Algorithm uses the initial connection weights and initial biases of the Long Short-Term Memory network as optimization objectives. Among them, the Grey Wolf optimization algorithm constructs a nonlinear convergence factor based on the hyperbolic tangent function during the iteration process, and dynamically adjusts the value of the nonlinear convergence factor according to the current iteration number. When updating the position of individuals in the population, the current positions of the multiple leader wolves with the best fitness values ​​are determined. The distance vector between the current positions of the multiple leader wolves and the current positions of the individuals in the population is calculated, and the positions of the individuals in the population are updated based on the distance vector and the nonlinear convergence factor.

[0011] Secondly, embodiments of this application provide a power plant data governance device, comprising: The data acquisition module is configured to acquire historical power plant data and real-time power plant data; The keyword extraction module is configured to extract keywords from historical power plant data and classify the historical power plant data according to power plant type based on the extracted keywords, thus obtaining the first historical power plant dataset divided by power plant type. The clustering module is configured to perform clustering on the first historical power plant dataset to obtain a second historical power plant dataset divided by business type under each power plant type. The data augmentation module is configured to augment the second historical power plant dataset to generate historical training sample sets for each business type. The model building module is configured to build a power plant data classification model using a deep learning model based on a historical training sample set; the initial network parameters of the deep learning model are determined by optimization using the Grey Wolf optimization algorithm. The classification processing module is configured to input real-time power plant data into the power plant data classification model, obtain classification results, and store the real-time power plant data based on the classification results.

[0012] In some embodiments, the data expansion module includes: The sample partitioning submodule is configured to divide the samples in the second historical power plant dataset into positive and negative samples based on the results of clustering. The positive sample augmentation submodule is configured to augment positive samples based on the normal distribution rule to obtain augmented positive samples; The negative sample augmentation submodule is configured to augment negative samples based on random distribution rules to obtain augmented negative samples. The sample combination submodule is configured to combine the augmented positive and negative samples to form a historical training sample set.

[0013] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the power plant data management methods.

[0014] Fourthly, embodiments of this application provide a computer program product, including a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any one of the power plant data governance methods.

[0015] The at least one technical solution adopted in this application can achieve the following beneficial effects: Pre-classification of historical power plant data through keyword extraction, dividing the historical power plant data according to power plant type, realizes hierarchical and precise division of multi-source heterogeneous power plant data, effectively breaking down data silos; Keyword extraction based on word frequency-inverse text frequency index-discrete factor algorithm and improved fuzzy C-means clustering algorithm based on adaptive fuzziness parameter and feature weighted distance metric fully explore the deep features of the data, effectively improving the classification accuracy of complex power plant data; Data augmentation by applying normal and random distribution rules to positive and negative samples respectively solves the sample imbalance problem, enhancing the diversity of training samples and the model's generalization ability; The gray wolf optimization algorithm is used to optimize the initial network parameters of the long short-term memory network, and through nonlinear convergence factor and multi-leader wolf collaborative guidance mechanism, effectively avoids the model training from getting trapped in local optima, improving the model's convergence speed and classification accuracy. The technical solution of this application effectively achieves efficient and unified governance of multi-type and multi-modal power plant data.

[0016] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description

[0017] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain the application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 This illustration schematically shows an overall process diagram of a power plant data governance method according to an embodiment of this application; Figure 2 This illustration schematically shows a keyword extraction process of a power plant data governance method according to an embodiment of this application; Figure 3 This illustration schematically shows a clustering process diagram of a power plant data governance method according to an embodiment of this application; Figure 4 This illustration schematically shows a data expansion process diagram of a power plant data governance method according to an embodiment of this application; Figure 5 This illustration schematically shows a process for optimizing the initial parameters of a deep learning model in a power plant data governance method according to an embodiment of this application. Figure 6 This illustration schematically shows a real-time power plant data classification process according to an embodiment of the power plant data governance method of this application; Figure 7 This illustration schematically shows an overall structural diagram of a power plant data management device according to an embodiment of this application; Figure 8 This illustration schematically shows a data expansion module structure of a power plant data management device according to an embodiment of this application; Figure 9 An exemplary block diagram of a computer program product of a power plant data management method according to an embodiment of this application is shown schematically.

[0019] In the diagram: 701, Data Acquisition Module; 702, Keyword Extraction Module; 703, Clustering Processing Module; 704, Data Expansion Module; 705, Model Building Module; 706, Classification Processing Module; 707, Sample Splitting Submodule; 708, Positive Sample Expansion Submodule; 709, Negative Sample Expansion Submodule; 710, Sample Combination Submodule; 901, Computer Program. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application more apparent, exemplary embodiments according to this application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0021] This application provides a power plant data governance method applicable to scenarios of unified management and intelligent governance of power plant data. For example, this method can be applied to heterogeneous data aggregation platforms for various types of power plants such as thermal power plants, wind farms, and nuclear power plants; centralized monitoring centers for regional power plant clusters; or data platform systems of power groups. In these scenarios, the data generated by each power plant comes from a wide range of sources and has diverse structures, including audio, images, and text, and suffers from problems such as data silos and inconsistent classification standards, making it difficult to achieve unified management and efficient utilization of data across power plants and business operations. The following, in conjunction with... Figure 1 This application provides a detailed description of a power plant data governance method based on an embodiment.

[0022] Figure 1 The illustration shows an overall process diagram of a power plant data governance method according to an embodiment of this application.

[0023] like Figure 1 As shown, the steps include: S101, Acquire historical power plant data. Specifically, historical power plant data comes from various sensors, smart meters, data acquisition terminals, and other devices, covering operational records under different power plant types and business scenarios, serving as the data foundation for subsequent data processing and model training.

[0024] S102, Extract keywords from historical power plant data. Based on the extracted keywords, classify the historical power plant data according to power plant type, obtaining the first historical power plant dataset divided by power plant type. Specifically, extract keywords from the acquired historical power plant data to identify keywords that can characterize the data source and content features. Based on the extracted keywords, pre-classify the historical power plant data according to power plant type, such as dividing it into thermal power plant data, wind farm data, nuclear power plant data, etc., obtaining the first historical power plant dataset divided by power plant type.

[0025] S103, perform clustering processing on the first historical power plant dataset to obtain a second historical power plant dataset for each power plant type, categorized by business type. Specifically, clustering processing is performed on the first historical power plant dataset corresponding to each power plant type. Through clustering analysis, the data under the same power plant type is further subdivided into different business types, such as equipment inspection data, power generation dispatch data, and fuel management data, thereby obtaining a second historical power plant dataset for each power plant type, categorized by business type.

[0026] S104, augment the second historical power plant dataset to generate historical training sample sets for each business type. Specifically, augmentation is performed on the second historical power plant dataset corresponding to each business type. This augmentation increases the diversity and quantity of training samples, resolving the issue of insufficient sample size for some business types, thereby generating historical training sample sets for each business type and providing sufficient and balanced training data for subsequent model training.

[0027] S105. Based on historical training sample sets, a power plant data classification model is constructed using a deep learning model. The initial network parameters of the deep learning model are determined through optimization using the Grey Wolf optimization algorithm. Specifically, the generated historical training sample set is used as training data to construct the power plant data classification model using a deep learning model. During model construction, the Grey Wolf optimization algorithm is used to optimize the initial network parameters of the deep learning model, obtaining the optimal initial network parameters as the starting point for model training to improve the model's convergence speed and classification accuracy.

[0028] S106: Input real-time power plant data into the power plant data classification model to obtain classification results, and store the real-time power plant data according to the classification results. Specifically, input the real-time collected power plant data into the constructed power plant data classification model, which classifies and identifies the real-time power plant data, outputs the corresponding classification results, and stores the real-time power plant data into the corresponding data categories according to the classification results, thus completing the classification and storage management of real-time power plant data.

[0029] Figure 2 The illustration shows a keyword extraction process of a power plant data governance method according to an embodiment of this application.

[0030] like Figure 2 As shown, the steps include: S201, Extract feature words from historical power plant data. Specifically, perform text parsing or feature recognition on the acquired historical power plant data files to extract words that can characterize the data content and attributes as candidate feature words. These feature words cover professional terms, equipment names, and business identifiers involved in power plant operation, ensuring that the extracted feature words can effectively reflect the semantic content of historical power plant data and provide a data object basis for subsequent calculation of the importance weight of feature words.

[0031] S202, based on the term frequency-inverse text frequency index-discrete factor algorithm, determines the weight of each feature word. Specifically, a keyword extraction model is constructed based on the term frequency-inverse text frequency index-discrete factor algorithm. This model is used to calculate the inter-class and intra-class discrete factors of each feature word to evaluate its distinguishing ability among different types of power plant data and its distribution within the same type of data. Simultaneously, the term frequency and inverse text frequency of each feature word in the current historical power plant data are statistically analyzed. Based on the inter-class discrete factor, intra-class discrete factor, term frequency, and inverse text frequency, the weight of each feature word is comprehensively calculated and determined to ensure that the weight accurately reflects the criticality and distinguishability of the feature word.

[0032] (Formula 1) For example, Formula 1 is used as the formula for calculating the standard deviation of feature words, where, Characteristic words Standard deviation; Characteristic words In category Frequency of occurrence in; Characteristic words Average frequency of occurrence across all categories; This represents the total number of data categories for power plants. This is a category indicator used to iterate through all data categories.

[0033] (Formula 2) For example, Formula 2 is used as the formula for calculating the inter-class scatter factor, where, Characteristic words The inter-class discrepancy factor is used to measure the ability of feature words to distinguish between different types of power plant data; Pi is a constant. It is the arctangent function; Characteristic words The standard deviation.

[0034] (Formula 3) For example, Formula 3 is used as the formula for calculating the intra-class feature dispersion, where, For category Chinese characteristic words The degree of dispersion; Characteristic words In category Frequency of occurrence in; For included categories The total number of historical power plant data files; For containing feature words And belong to the category The number of historical power plant data files.

[0035] (Formula 4) For example, Formula 4 is used as the formula for calculating the within-class discrete factor, where, Characteristic words The intra-class discrete factor is used to measure the uniformity of the distribution of feature words within the same type of power plant data; Pi is a constant. It is the arctangent function; For category Chinese characteristic words The degree of dispersion.

[0036] (Formula 5) For example, Formula 5 is used as the formula for calculating feature word weights, where, Characteristic words The final weight; The frequency of the characteristic words; The inverse text frequency of the feature words; Characteristic words In category Intra-class discrete factor; Characteristic words The inter-class discrete factor.

[0037] S203, sort the feature words according to their weights. Specifically, the calculated weights of each feature word are used as the sorting index, and all candidate feature words are arranged in descending order of weight. Through this sorting operation, the feature words that are most representative and discriminative for the current historical power plant data are selected.

[0038] S204, Select a preset number of feature words as keywords for historical power plant data. Specifically, according to the preset selection quantity, select a corresponding number of feature words from the sorted feature word sequence and determine them as keywords for historical power plant data, which will be used for subsequent classification of historical power plant data by power plant type.

[0039] Figure 3The illustration shows a clustering process diagram of a power plant data governance method according to an embodiment of this application.

[0040] like Figure 3 As shown, the steps include: S301, Initialize the membership matrix of the fuzzy C-means clustering algorithm. Specifically, for the first historical power plant dataset corresponding to each power plant type, initialize the membership matrix of the fuzzy C-means clustering algorithm. Each element in the membership matrix represents the degree of membership of the sample data point to each cluster center, and its value ranges from 0 to 1. It is initialized randomly or based on prior knowledge.

[0041] S302, based on the current membership matrix, calculate the cluster centers using an adaptive ambiguity parameter and a feature-weighted distance metric. Specifically, the adaptive ambiguity parameter is determined according to the current iteration round. This parameter has a larger value in the early stages of iteration, allowing samples to belong to multiple categories with higher ambiguity, which is beneficial for global exploration; its value decreases in the later stages of iteration, making the clustering results more explicit. Simultaneously, when calculating the distance from a sample to the cluster center, a feature-weighted distance metric is introduced, assigning different weights according to the importance of each feature dimension, with key features having a greater weight in the distance calculation. Based on the current membership matrix, the adaptive ambiguity parameter, and the feature-weighted distance metric, calculate and update the cluster center positions corresponding to each business type.

[0042] (Formula 6) For example, Formula 6 is used as the formula for calculating the adaptive ambiguity parameter, where, For the first The ambiguity parameter at the next iteration; , These are the preset maximum and minimum ambiguities, respectively; This represents the current iteration number; This represents the maximum number of iterations for the clustering algorithm.

[0043] (Formula 7) For example, Formula 7 is used as the formula for calculating the feature-weighted distance, where, For the sample To the cluster center The weighted distance; Total dimension of data features; For the first Weights of dimensional features; For the sample In the eigenvalues ​​of dimension; Cluster center In the Eigenvalues ​​of dimension.

[0044] S303: Update the membership matrix based on the current cluster centers and calculate the objective function value. Specifically, based on the newly calculated cluster centers, recalculate the membership degree of each sample data point to each cluster center and update the membership matrix. Simultaneously, calculate the objective function value for the current iteration. This objective function value reflects the sum of weighted distances from all sample data points to their respective cluster centers and is used to measure the quality of the current clustering effect; a smaller objective function value indicates a better clustering effect.

[0045] (Formula 8) For example, Formula 8 is used as the formula for calculating cluster centers, where, For the first Cluster centers; For data points Belongs to the The membership degree of each cluster; For ambiguity parameters; For the first Historical data sample of a power plant; This represents the total number of historical data samples for the current power plant type.

[0046] (Formula 9) For example, Formula 9 is used as the formula for calculating membership degree, where, For data points Belongs to the The membership degree of each cluster, with values ​​ranging from [0,1]; , The first The sample to the first , The distance between the cluster centers; This represents the total number of cluster centers. This is the ambiguity parameter.

[0047] (Formula 10) For example, Equation 10 is used as the clustering objective function, where, This represents the objective function value for FCM clustering; the smaller the value, the better the clustering effect. The membership degree of the sample; For ambiguity parameters; For sample data; As cluster center; The total number of samples; This represents the total number of cluster centers.

[0048] S304, determine whether the clustering meets the convergence criteria. Specifically, determine whether the current clustering meets the preset convergence criteria. Convergence criteria include, but are not limited to, the change in the objective function value between two consecutive iterations being less than a preset threshold, or the number of iterations reaching the preset maximum number of iterations. If the convergence criteria are met, stop the iteration and output the second historical power plant dataset divided by business type for each power plant type; if the convergence criteria are not met, return to step S302 to continue the next round of iteration updates until the convergence criteria are met.

[0049] (Formula 11) For example, Formula 11 is used as the convergence criterion formula, where, For the first The objective function value of the next iteration; For the first The objective function value of the next iteration; The preset convergence threshold is used; if this formula is satisfied, the clustering algorithm is considered to have converged.

[0050] Figure 4 The illustration shows a schematic diagram of the data expansion process of a power plant data governance method according to an embodiment of this application.

[0051] like Figure 4 As shown, the steps include: S401, Based on the clustering results, the samples in the second historical power plant dataset are divided into positive and negative samples. Specifically, for the second historical power plant dataset corresponding to each business type, the samples are divided according to the clustering results. The clustering results include the membership information between the samples and each cluster center. Based on this membership information, typical samples belonging to the current business type are classified as positive samples, and samples that do not belong to the current business type or are in the boundary region are classified as negative samples.

[0052] S402, augment the positive samples based on the normal distribution rule to obtain augmented positive samples. Specifically, for the defined positive samples, data augmentation is performed based on the normal distribution rule. For each positive sample, the mean and standard deviation of its numerical feature vector are calculated. Based on this, new samples are generated according to the normal distribution rule, so that the newly generated samples retain the core features of the original samples while introducing small differences, thereby increasing the diversity and quantity of positive samples, resulting in augmented positive samples.

[0053] S403, data augmentation is performed on the negative samples based on random distribution rules to obtain augmented negative samples. Specifically, for the defined negative samples, data augmentation is performed based on random distribution rules. The global distribution range of all negative samples on numerical features is statistically analyzed, and new samples are generated within this range according to random distribution rules. This allows the newly generated negative samples to cover a wider feature space, enhancing the model's ability to identify outlier and boundary data, resulting in augmented negative samples.

[0054] S404 combines the expanded positive and negative samples to form a historical training sample set. Specifically, the positive and negative samples obtained after data augmentation are integrated to form historical training sample sets for each business type. This historical training sample set contains sufficient and balanced positive and negative samples, providing a high-quality data foundation for subsequent deep learning model training and helping to improve the model's classification accuracy and generalization ability.

[0055] Figure 5 The illustration shows a schematic diagram of the initial parameter optimization process of a deep learning model for a power plant data governance method according to an embodiment of this application.

[0056] like Figure 5 As shown, the steps include: S501, Initialize the gray wolf population by using the initial connection weights and initial biases of the Long Short-Term Memory (LSTM) network as the position vectors of individual gray wolves. Specifically, determine the population size for the gray wolf optimization algorithm, initialize the gray wolf population, and each gray wolf in the population represents a set of candidate initial parameters for the LSM network. Encode the initial connection weights and initial biases of the LSM network into position vectors for individual gray wolves, with the dimension of each gray wolf's position vector matching the total number of parameters in the LSM network. Generate the initial population through random initialization to provide starting search points for the subsequent optimization process.

[0057] S502, calculate the fitness value of each gray wolf individual and determine the multiple leader wolves with the best fitness values. Specifically, for each gray wolf individual in the population, its corresponding initial connection weights and initial biases are used as the initial parameters of the Long Short-Term Memory network. The network is trained based on the historical training sample set, and the prediction error on the validation set is calculated. The reciprocal or negative value of the prediction error is used as the fitness value of that gray wolf individual; a higher fitness value indicates better initial parameters. All gray wolves in the population are sorted according to the calculated fitness values, and the multiple individuals with the best fitness values ​​are determined as leader wolves. These leader wolves guide the position updates of other gray wolves in subsequent iterations.

[0058] (Formula 12) For example, Equation 12 is used as the fitness function, where, For individual gray wolves fitness value; To predict the mean square error; The total number of samples in the validation set; These are the model's predicted values; This represents the actual data value.

[0059] In S503, during the iteration process, a nonlinear convergence factor is constructed based on the hyperbolic tangent function, and its value is dynamically adjusted according to the current iteration number. Specifically, in the iteration process of the Grey Wolf optimization algorithm, a nonlinear convergence factor is introduced to balance the algorithm's global exploration capability and local exploitation capability. The nonlinear convergence factor, constructed based on the hyperbolic tangent function, exhibits a nonlinear decay trend with increasing iteration number. In the early stages of iteration, the convergence factor has a larger value, which is beneficial for the algorithm to broadly search for the optimal solution globally; in the later stages of iteration, the convergence factor gradually decreases, enabling the algorithm to perform fine-grained searches in local regions and improve optimization accuracy.

[0060] (Formula 13) For example, Equation 13 is used as the formula for calculating the nonlinear convergence factor, where, It is a nonlinear convergence factor; It is the hyperbolic tangent function; This represents the current iteration number; This is the threshold for the number of iterations; The deceleration rate parameter; The periodic parameter is decreasing. , These are the maximum and minimum values ​​of the convergence factor, respectively.

[0061] S504: A distance vector is calculated based on the current positions of multiple leader wolves and the current position of the individual gray wolf. The position of the individual gray wolf is then updated based on the distance vector and a nonlinear convergence factor. Specifically, for each individual gray wolf in the population, the distance vector between it and each leader wolf is calculated. This distance vector reflects the difference between the current gray wolf and the leader wolves in the parameter space. Based on the calculated distance vector and the nonlinear convergence factor corresponding to the current iteration, the position update direction of the current gray wolf under the guidance of each leader wolf is determined. A new position vector is generated by integrating the guidance information from multiple leader wolves, completing the update of the gray wolf's position and causing the population to gradually cluster towards parameter regions with higher fitness values.

[0062] (Formula 14) For example, Formula 14 is used as the enclosing action formula, where, , For the first , The individual position of the gray wolf at any given moment; Location of the prey; The distance between the gray wolf and its prey; This is the convergence influence coefficient vector; The vector of oscillation influence coefficients; It is a nonlinear convergence factor; , It is a random number within the interval [0,1].

[0063] (Formula 15) For example, Formula 15 is used as the hunting action formula, where, , , The optimal ones are respectively , , The position of the leader of the wolves; , , The updated position is guided by the leader wolf; , , , , , This represents the influence coefficient of the corresponding leader, the wolf. , , This refers to the distance between an ordinary gray wolf and a alpha wolf.

[0064] (Formula 16) For example, Formula 16 is used as the position update formula, where, The updated position of the individual gray wolf; , , They are respectively , , The candidate positions, guided by the leader wolf, are updated by taking the average of the three factors.

[0065] S505, determine whether the preset maximum number of iterations or convergence condition has been reached. Specifically, after each iteration update, determine whether the current iteration termination condition is met. The termination condition includes the current iteration count reaching the preset maximum number of iterations, or the position change of the gray wolf with the best fitness value in multiple consecutive iterations being less than a preset threshold. If the termination condition is met, stop the iteration, output the position vector of the gray wolf with the best fitness value, and the initial connection weights and initial biases corresponding to this position vector are the optimal initial parameters of the Long Short-Term Memory network; if the termination condition is not met, return to step S502 to continue the next round of iteration until the termination condition is met.

[0066] Figure 6 The illustration shows a real-time power plant data classification process of a power plant data governance method according to an embodiment of this application.

[0067] like Figure 6 As shown, the steps include: S601, Acquire real-time power plant data. Specifically, collect various types of data in real time during the power plant's production and operation, including sensor data, smart meter readings, monitoring images, equipment operation logs, and other modalities of data. This real-time power plant data serves as input data for classification, reflecting the current operating status of the power plant and the condition of its equipment, providing a data foundation for subsequent classification and identification.

[0068] S602, input real-time power plant data into the power plant data classification model. Specifically, the acquired real-time power plant data is preprocessed and then input into a pre-built power plant data classification model. This power plant data classification model is a classification model trained based on historical training sample sets, using a deep learning model and optimizing the initial network parameters using the Grey Wolf optimization algorithm. It can accurately classify the input power plant data.

[0069] S603 classifies real-time power plant data using a power plant data classification model and outputs the classification results. Specifically, the power plant data classification model extracts features and classifies the input real-time power plant data. Based on the content and attribute features of the data, it determines the power plant type and business type to which the real-time power plant data belongs and outputs the corresponding classification results.

[0070] S604. Store real-time power plant data based on the classification results. Specifically, based on the classification results output by the power plant data classification model, store the real-time power plant data into the corresponding data categories, completing the classification and storage management of real-time power plant data. This achieves unified classification and orderly storage of real-time power plant data, facilitating subsequent data retrieval, analysis, and application.

[0071] Figure 7 The schematic diagram illustrates the overall structure of a power plant data management device according to an embodiment of this application.

[0072] like Figure 7 As shown, the device includes a data acquisition module 701, a keyword extraction module 702, a clustering processing module 703, a data expansion module 704, a model building module 705, and a classification processing module 706.

[0073] The data acquisition module 701 is configured to acquire historical power plant data and real-time power plant data. The historical power plant data comes from various field devices such as sensors, smart meters, and data acquisition terminals, covering operation record data of different power plant types and different business scenarios; the real-time power plant data is multimodal data collected in real time during the power plant's production and operation, including sensor data, smart meter readings, monitoring images, and equipment operation logs.

[0074] The keyword extraction module 702 is configured to extract keywords from historical power plant data and classify the historical power plant data according to the power plant type based on the extracted keywords, such as thermal power plant data, wind farm data, nuclear power plant data, etc., to obtain the first historical power plant dataset classified by power plant type.

[0075] The clustering module 703 is configured to perform clustering processing on the first historical power plant dataset to obtain a second historical power plant dataset divided by business type for each power plant type. Specifically, the clustering module 703 performs clustering processing on the first historical power plant dataset corresponding to each power plant type. Through clustering analysis, the data under the same power plant type is further subdivided into different business types, such as equipment inspection data, power generation dispatch data, fuel management data, etc., thereby obtaining a second historical power plant dataset divided by business type for each power plant type.

[0076] The data augmentation module 704 is configured to augment the second historical power plant dataset to generate historical training sample sets for each business type. Specifically, the data augmentation module 704 performs data augmentation processing on the second historical power plant dataset corresponding to each business type. Through data augmentation, the diversity and quantity of training samples are increased, addressing the problem of insufficient sample size for some business types, thereby generating historical training sample sets for each business type and providing sufficient and balanced training data for subsequent model training.

[0077] The model building module 705 is configured to construct a power plant data classification model using a deep learning model based on a historical training sample set. Specifically, the initial network parameters of the deep learning model are optimized using the Grey Wolf optimization algorithm. The model building module 705 uses the generated historical training sample set as training data to construct the power plant data classification model using a deep learning model. During the model building process, the Grey Wolf optimization algorithm is used to optimize the initial network parameters of the deep learning model, obtaining the optimal initial network parameters as the starting point for model training to improve the model's convergence speed and classification accuracy.

[0078] The classification processing module 706 is configured to input real-time power plant data into the power plant data classification model, obtain classification results, and store the real-time power plant data according to the classification results, thereby completing the classification and storage management of real-time power plant data.

[0079] Figure 8 The illustration shows a schematic diagram of the data expansion module structure of a power plant data management device according to an embodiment of this application.

[0080] like Figure 8 As shown, the data expansion module 704 includes a sample partitioning submodule 707, a positive sample expansion submodule 708, a negative sample expansion submodule 709, and a sample combination submodule 710.

[0081] The sample partitioning submodule 707 is configured to divide the samples in the second historical power plant dataset into positive and negative samples based on the results of clustering. Specifically, the sample partitioning submodule 707 partitions the samples in the second historical power plant dataset corresponding to each business type based on the results of clustering. The results of clustering include the membership information between the samples and each cluster center. Based on this membership information, samples belonging to the current business type are classified as positive samples, and samples not belonging to the current business type are classified as negative samples.

[0082] The positive sample expansion submodule 708 is configured to expand the positive samples based on the normal distribution rule to obtain expanded positive samples. Specifically, the positive sample expansion submodule 708 expands the data of the divided positive samples based on the normal distribution rule. For each positive sample, the mean and standard deviation of its numerical feature vector are calculated. Based on this, new samples are generated according to the normal distribution rule, so that the newly generated samples retain the core features of the original samples while introducing small differences, thereby increasing the diversity and quantity of positive samples, resulting in expanded positive samples.

[0083] The negative sample augmentation submodule 709 is configured to augment negative samples based on random distribution rules to obtain augmented negative samples. Specifically, the negative sample augmentation submodule 709 augments the defined negative samples based on random distribution rules. It calculates the global distribution range of all negative samples across numerical features, and generates new samples within this range according to random distribution rules. This allows the newly generated negative samples to cover a wider feature space, enhancing the model's ability to identify outlier and boundary data, thus obtaining the augmented negative samples.

[0084] The sample combination submodule 710 is configured to combine the augmented positive and negative samples to form a historical training sample set. Specifically, the sample combination submodule 710 integrates the augmented positive and negative samples to form historical training sample sets for each business type. This historical training sample set contains sufficient and balanced positive and negative samples, providing a high-quality data foundation for the subsequent training of deep learning models and helping to improve the classification accuracy and generalization ability of the models.

[0085] Figure 9 An exemplary block diagram of a computer program product of a power plant data management method according to an embodiment of this application is shown schematically.

[0086] like Figure 9 As shown, the computer program product stores a computer program 901, which, when executed by a processor, implements the method provided in any embodiment of this application.

[0087] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0088] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0089] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0090] It should also be noted that in the system and method of this application, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions of this application.

[0091] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this application is not limited to the specific aspects of the processes, machines, manufacturing, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufacturing, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described above can be utilized. Therefore, the appended claims include such processes, machines, manufacturing, events, means, methods, or actions within their scope.

[0092] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0093] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A power plant data governance method, characterized in that, include: Obtain historical power plant data; Keyword extraction is performed on the historical power plant data, and the historical power plant data is classified according to the power plant type based on the extracted keywords to obtain the first historical power plant dataset divided by the power plant type. Cluster the first historical power plant dataset to obtain a second historical power plant dataset divided by business type for each power plant type. The second historical power plant dataset is augmented to generate historical training sample sets for each of the aforementioned business types; Based on the historical training sample set, a power plant data classification model is constructed using a deep learning model; wherein, the initial network parameters of the deep learning model are determined by optimization using the Grey Wolf optimization algorithm. Real-time power plant data is input into the power plant data classification model to obtain classification results, and the real-time power plant data is stored based on the classification results.

2. The power plant data governance method according to claim 1, characterized in that, The historical power plant data includes at least one of audio data, image data, and character data.

3. The power plant data governance method according to claim 1, characterized in that, Keyword extraction was performed on the historical power plant data, including: Based on the word frequency-inverse text frequency index-discrete factor algorithm, feature words are extracted from the historical power plant data, and the weight of each feature word is determined. The feature words are sorted according to the weights, and a preset number of the feature words are selected as the keywords of the historical power plant data.

4. The power plant data governance method according to claim 1, characterized in that, The clustering process is based on the fuzzy C-means clustering algorithm and performs clustering based on adaptive fuzziness parameters and feature-weighted distance metrics.

5. The power plant data governance method according to claim 1, characterized in that, The data augmentation of the second historical power plant dataset includes: Based on the results of clustering, the samples in the second historical power plant dataset are divided into positive samples and negative samples; The positive samples are augmented based on the normal distribution rule to obtain augmented positive samples; The negative samples are augmented based on random distribution rules to obtain augmented negative samples; The augmented positive samples and negative samples are combined to form the historical training sample set.

6. The power plant data governance method according to claim 1, characterized in that, The deep learning model employs a long short-term memory network; The Grey Wolf Optimization Algorithm uses the initial connection weights and initial biases of the Long Short-Term Memory network as optimization objectives. In the process of iteration, the gray wolf optimization algorithm constructs a nonlinear convergence factor based on the hyperbolic tangent function and dynamically adjusts the value of the nonlinear convergence factor according to the current iteration number. When updating the position of individuals in the population, the current positions of the multiple leader wolves with the best fitness values ​​are determined, a distance vector is calculated based on the current positions of the multiple leader wolves and the current positions of the individuals in the population, and the position of the individuals in the population is updated based on the distance vector and the nonlinear convergence factor.

7. A power plant data management device, characterized in that, include: The data acquisition module is configured to acquire historical power plant data and real-time power plant data; The keyword extraction module is configured to extract keywords from the historical power plant data and classify the historical power plant data according to the power plant type based on the extracted keywords, thereby obtaining a first historical power plant dataset divided according to the power plant type. The clustering processing module is configured to perform clustering processing on the first historical power plant dataset to obtain a second historical power plant dataset divided by business type under each power plant type; The data augmentation module is configured to augment the second historical power plant dataset to generate historical training sample sets for each of the aforementioned business types. The model building module is configured to build a power plant data classification model using a deep learning model based on the historical training sample set; wherein, the initial network parameters of the deep learning model are determined by optimization using the Grey Wolf optimization algorithm. The classification processing module is configured to input the real-time power plant data into the power plant data classification model, obtain the classification result, and store the real-time power plant data according to the classification result.

8. The power plant data governance device according to claim 7, characterized in that, The data expansion module includes: The sample partitioning submodule is configured to divide the samples in the second historical power plant dataset into positive samples and negative samples based on the results of clustering. The positive sample augmentation submodule is configured to augment the positive sample based on the normal distribution rule to obtain the augmented positive sample; The negative sample augmentation submodule is configured to augment the negative sample based on a random distribution rule to obtain the augmented negative sample. The sample combination submodule is configured to combine the augmented positive samples and the negative samples to form the historical training sample set.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the power plant data management method according to any one of claims 1 to 6.

10. A computer program product comprising a computer-readable storage medium on which a computer program is stored, characterized in that, When the computer program is executed by the processor, it implements the power plant data management method according to any one of claims 1 to 6.

Images