Health assessment method and device of power equipment, computer equipment, readable storage medium and program product

Abstract

The invention relates to a health assessment method and device for power equipment, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring real-time power data of each power device in a target power grid system; the real-time power data comprises a basic evaluation index; calculating a health degree score of each basic evaluation index according to a pre-calculated health degree function; the health degree function is calculated according to each health sample data; constructing a knowledge graph based on the health degree scores, and predicting health degree scores of the power devices in a future time period based on the knowledge graph and target weights corresponding to the power devices; the target weight is obtained by adjusting the target weight corresponding to the previous moment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid systems, in particular to a health assessment method and device for power equipment, computer equipment, computer readable storage medium and computer program product. BACKGROUND

[0002] With the continuous expansion of the power system scale and the increasingly complex equipment operating environment, online monitoring technology has been widely applied. By collecting multi-source operating data such as electrical parameters, thermal parameters, insulation state, mechanical vibration, etc., the state of the power equipment can be judged and operation and maintenance decisions can be made. At present, the common health assessment method for power equipment is usually based on threshold judgment, linear weighted scoring or historical data regression, etc. to quantify the operating state of the equipment.

[0003] In the traditional technology, the collected monitoring data is usually compared with the preset threshold, or a plurality of indexes are weighted calculated by fixed weight to obtain the health score of the equipment, which is used to reflect the current operating state.

[0004] However, these methods mostly rely on static threshold or fixed scoring model, and fail to effectively reflect the device operating conditions, index correlation and data distribution changes over time. Therefore, the health score obtained by the above method is often not accurate enough. SUMMARY

[0005] Therefore, it is necessary to provide a health assessment method, device, computer equipment, computer readable storage medium and computer program product for power equipment capable of predicting health score in view of the above technical problems.

[0006] In a first aspect, the present application further provides a health assessment method for power equipment, comprising:

[0007] Obtaining real-time power data of each power equipment in a target power grid system; the real-time power data includes basic evaluation indexes;

[0008] Calculating the health score of each basic evaluation index according to a pre-calculated health function; the health function is calculated according to each health sample data;

[0009] Based on the health score, a knowledge graph is constructed, and based on the knowledge graph and the target weight corresponding to each power equipment, the health score of each power equipment in the future time period is predicted; the target weight is obtained by adjusting the target weight corresponding to the last time.

[0010] In one of the embodiments, the knowledge graph comprises nodes and edges between the nodes; the nodes are basic evaluation indexes; the edges are determined according to the correlations between the basic evaluation indexes; and the node features of the nodes are health scores;

[0011] The health scores of the power equipment in the future time period are predicted based on the knowledge graph and target weights corresponding to the power equipment, comprising:

[0012] The knowledge graph and the target weights are input into a health evaluation model to obtain a health time series curve corresponding to the current time;

[0013] The health scores of the power equipment in the future time period are predicted based on the knowledge graph and target weights corresponding to the power equipment, comprising:

[0014] In one of the embodiments, the method further comprises:

[0015] The health scores of the power equipment in the future time period are calculated with a slope and a segmented slope; the slope is used to represent the change trend of the health score in the future time period; and the segmented slope is used to represent the local change trend of the health score in each sub time period;

[0016] The slope and the segmented slope are compared with a preset threshold value respectively to obtain a future change result.

[0017] In one of the embodiments, the threshold value comprises an upper limit threshold value and a safety critical value; the upper limit threshold value is determined according to the statistical upper limit value of the historical health score distribution of the power equipment in the normal operation state; and the safety critical value is a critical value determined based on the equipment safety operation margin.

[0018] In one of the embodiments, the calculation method of the health function comprises:

[0019] The kernel density estimation is performed on the health sample data to obtain a target bandwidth;

[0020] The target bandwidth and a Gaussian kernel function are used for calculation to obtain the probability density under different values;

[0021] The probability density is normalized to obtain the health function.

[0022] In one of the embodiments, the adjustment process of the target weight comprises:

[0023] The target weight of the last time and the health scores of the corresponding basic evaluation indexes are input into a state input weight adjustment model;

[0024] Generate the corresponding weight adjustment amount based on the weight adjustment model;

[0025] The target weight at the previous time step is updated based on the weight adjustment amount to obtain the adjusted target weight.

[0026] Secondly, this application also provides a health assessment device for electrical equipment, comprising:

[0027] The acquisition module is used to acquire real-time power data of each power device in the target power grid system;

[0028] The health calculation module is used to calculate the health score of each of the real-time power data according to a pre-calculated health function; the health function is calculated based on each health sample data; the basic evaluation index is determined based on the physical structure and working principle of each of the power devices.

[0029] The health prediction module is used to construct a knowledge graph based on the health score, and predict the health score of each power device in a future time period based on the knowledge graph and the target weights corresponding to each power device; the target weights are obtained by adjusting the target weights corresponding to the previous time.

[0030] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method in any of the above embodiments.

[0031] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method in any of the above embodiments.

[0032] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the methods in any of the above embodiments.

[0033] The aforementioned health assessment methods, devices, computer equipment, computer-readable storage media, and computer program products for power equipment utilize a health function constructed based on health sample data. This function represents the mapping relationship between the values ​​of basic assessment indicators and the health status, enabling the conversion of real-time power data into health scores characterizing the current operating state of the power equipment. Subsequently, a knowledge graph is constructed based on the health scores to model the relationships between various basic assessment indicators. A comprehensive evaluation is then performed using target weights to obtain the current health score for each power device. Furthermore, by generating time-series data from the health scores and using this data for prediction, the health scores for each power device in future time periods can be obtained. Thus, by combining the health function, the knowledge graph, and time-series prediction based on historical states, the future health status of power equipment can be predicted more accurately. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart illustrating a method for accurately predicting the future health status of electrical equipment in one embodiment.

[0036] Figure 2 A structural block diagram of a device for accurately predicting the future health status of electrical equipment in one embodiment;

[0037] Figure 3 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0039] In one embodiment, such as Figure 1 As shown, a health assessment method for power equipment is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0040] Step 102: Obtain real-time power data of each power device in the target power grid system; real-time power data includes basic evaluation indicators.

[0041] Optionally, the status monitoring devices installed on each power equipment collect equipment operating parameters and upload the operating parameters to the data management platform through a communication network; wherein the status monitoring devices include sensor acquisition modules, intelligent acquisition terminals or online detection devices, and the communication network includes fiber optic communication networks, wireless private networks or industrial Ethernet.

[0042] For example, real-time power data may include voltage, current, power factor, active power, reactive power, etc., which characterize electrical operating status; temperature and temperature rise rate, which characterize thermal status; dielectric loss, leakage current, partial discharge, etc., which characterize insulation status; and operating parameters such as vibration, stroke, and number of actions, which characterize mechanical characteristics.

[0043] Optionally, the real-time power data of each power device corresponds to multiple dimensions, each dimension includes at least one performance indicator, and each performance indicator corresponds to a basic evaluation indicator.

[0044] For example, power data includes common health dimensions, which at least include: insulation performance, electrical performance, thermal performance, mechanical structure, and environmental adaptability. The basic evaluation indicators for insulation performance at least include: insulation resistance, dielectric loss factor, and partial discharge. The basic evaluation indicators for electrical performance at least include: operating parameter deviation, i.e., voltage / current deviation from rated values ​​and grounding resistance. The basic evaluation indicators for thermal performance at least include: operating temperature and temperature change rate. The basic evaluation indicators for mechanical / structural performance at least include: sealing performance, i.e., leakage rate, degree of damage, and structural vibration amplitude. The basic evaluation indicators for environmental adaptability at least include: temperature and humidity deviation, corrosive gas concentration, and condensation risk index.

[0045] Step 104: Calculate the health score of each basic assessment indicator based on the pre-calculated health score function; the health score function is calculated based on the health sample data.

[0046] Among them, the health function is used to characterize the correspondence between the values ​​of the basic assessment indicators and the health status, and is used to map the real-time monitoring values ​​into a health score in the range of [0,1]. The closer the health score is to 1, the closer the operating status is to a healthy condition.

[0047] For example, the current values ​​of each basic assessment indicator in the real-time power data can be used as input to the health function to obtain the corresponding health score.

[0048] Furthermore, basic evaluation indicators corresponding to the target equipment can be filtered from real-time power data, and these indicators can be input into a health function to obtain a health score for each indicator. By calculating the health score only for the basic evaluation indicators, invalid data can be avoided from being introduced into the calculation process, thus improving the accuracy and efficiency of the health score.

[0049] Furthermore, the method for constructing the health degree function includes: estimating the kernel density of the health sample data to determine the target bandwidth used to characterize the probability density distribution; calculating the density of the health sample data based on the target bandwidth and the kernel function to obtain the probability density of the basic evaluation index under different values; normalizing the probability density to construct the health degree function of the basic evaluation index.

[0050] The health sample data consists of monitoring data collected from each power device under normal operating conditions, excluding data related to fault conditions, maintenance, or obvious abnormalities. For example, operating data of the power device within a preset time period after commissioning, provided there are no fault records, can be collected as the health sample data.

[0051] Optionally, the target bandwidth is used as a parameter to characterize the smoothness in kernel density estimation, aiming to achieve a balance between noise suppression and detail preservation. If the bandwidth is too small, the density function will overfit and fail to reflect the actual distribution within the normal operating range; if the bandwidth is too large, the health distribution may be overly smoothed, resulting in the loss of information about early signs of anomalies. Therefore, in order to make the health function more accurately reflect the distribution pattern of the basic assessment indicators under healthy conditions, the target bandwidth can be the optimal bandwidth describing the distribution characteristics of healthy sample data. For example, the expression for calculating the optimal bandwidth of kernel density estimation is shown in formula (1).

[0052] Formula (1)

[0053] In the formula, The optimal bandwidth for kernel density estimation. The standard deviation of the health sample set indicator. The sample size for the health sample set indicators;

[0054] According to the kernel density estimation formula, the probability density expression for each index at any value is calculated as shown in formula (2):

[0055] Formula (2)

[0056] In the formula, For the indicator at any point The probability density at that location, For standard Gaussian kernel function, For the first health sample set The value of a specific data point The values ​​of the indicators to be evaluated;

[0057] The health function expression is:

[0058] Formula (3)

[0059] In the formula, In order to be in The health score of the area probability density The maximum value within the numerical range represented by the healthy sample set.

[0060] Step 106: Construct a knowledge graph based on the health score, and predict the health score of each power device in the future time period based on the knowledge graph and the target weights corresponding to each power device; the target weights are obtained by adjusting the target weights corresponding to the previous time.

[0061] The target weight is used to characterize the contribution of each basic assessment indicator to the overall health status of power equipment.

[0062] Optionally, constructing a knowledge graph includes: identifying each basic assessment indicator as a graph node; calculating the correlation between indicators based on the joint distribution characteristics of the basic assessment indicators in historical health sample data; establishing edges between different basic assessment indicators based on the correlation; and using the health score corresponding to each basic assessment indicator as a node feature, so that the knowledge graph can represent the relationship between the basic assessment indicators and their health status.

[0063] Optionally, predicting the health score of power equipment in a future time period includes: obtaining the current health score of the power equipment based on a knowledge graph and target weights; forming a health time series of the health scores; inputting the health time series into the prediction model and outputting the health score of the power equipment in a preset future time period.

[0064] Furthermore, the prediction model outputs predicted health values ​​for multiple future time periods based on the health time series, and performs fusion calculations based on the predicted values ​​and target weights to reflect the degree of influence of different basic assessment indicators on future health status.

[0065] In this embodiment, by constructing a knowledge graph, multiple basic assessment indicators can be jointly modeled based on the relationships between nodes. This allows health status assessment to not only consider the health changes of individual indicators but also reflect the synergistic relationship between multiple indicators, thereby improving the accuracy and reliability of health prediction.

[0066] In the above embodiments, since the health function is constructed based on health sample data and is used to characterize the mapping relationship between the values ​​of basic assessment indicators and health status, real-time power data can be converted into a health score representing the current operating status of power equipment based on the health function. Then, a knowledge graph is constructed based on the health score to model the correlation between various basic assessment indicators, and a comprehensive evaluation is performed in conjunction with target weights to obtain the current health score of each power device. Based on this, by forming time-series data from the health scores and making predictions based on this time-series data, the health score of each power device in the future time period can be obtained. Thus, by combining the health function, the knowledge graph, and time-series prediction based on historical states, the future health status of power equipment can be predicted more accurately.

[0067] In one embodiment, the above-mentioned prediction of the health score of each power device in a future time period based on the knowledge graph and the target weights corresponding to each power device includes: inputting the knowledge graph and the target weights into the health assessment model to obtain the health time series curve corresponding to the current moment; and making predictions based on the health time series curve to obtain the health score of each power device in a future time period.

[0068] Among them, the health assessment model is a pre-trained health assessment model for processing graph-structured data. It can be built based on graph convolutional networks and is used to obtain the current health score of power equipment according to the correlation between various basic assessment indicators in the knowledge graph and the target weight.

[0069] Among them, the health assessment model is a pre-trained health assessment model for processing graph-structured data. It can be built based on graph convolutional networks and is used to obtain the current health score of power equipment according to the correlation between various basic assessment indicators in the knowledge graph and the target weight.

[0070] Optionally, obtaining the time series sequence of health scores includes: using the health scores of the basic assessment indicators as node features; performing multi-layer graph convolution calculations on the node features to obtain multi-layer aggregated node features; and performing global aggregation on the node features to obtain the health scores for power equipment.

[0071] Optionally, predicting the health score of each power device in the future time period includes: inputting the time series sequence of health scores into the time series prediction model, and outputting the target health score in the preset future time period.

[0072] For example, when constructing a knowledge graph for power equipment, the health score corresponding to the basic assessment indicators is used as the node feature. For any node corresponding to a basic assessment indicator, the node feature of that node in the previous layer is aggregated with the node features of its neighboring nodes to obtain the corresponding aggregated feature. The node feature and the aggregated feature are concatenated to form a joint feature vector. Based on a learnable transformation matrix and a nonlinear activation function, the joint feature vector is transformed to obtain the node feature of that node in the current layer. By repeating the above calculation process, the graph convolution calculation of the preset number of layers is completed to obtain the node features corresponding to all basic assessment indicators. After completing the graph convolution calculation, the node features are globally pooled to obtain the graph-level representation vector of the knowledge graph. The graph-level representation vector is input into the prediction model, and the health score of the power equipment at the current time is obtained through regression calculation. It can be explained that by using the health score of the basic assessment indicators as the node feature and by using a graph-based feature aggregation method, nonlinear correlation features can be learned from the basic assessment indicators and their relationships, avoiding linear weighted summarization based on a preset dimension division method, thus more accurately reflecting the impact of multi-indicator coupling on the health status of power equipment.

[0073] Furthermore, the health scores corresponding to the basic assessment indicators can be used to provide sub-item diagnostic results, while the graph-level representation vectors are used to generate system-level health scores. The two complement each other, providing both an overall operational status and a fine-grained indicator-level interpretation for anomaly localization.

[0074] For example, the health scores of power equipment at multiple consecutive times are used as the input sequence. The input sequence is fed into the time series prediction model, which outputs the health scores corresponding to the preset future time points. The model is then based on the health scores corresponding to the future time points.

[0075] In one embodiment, the training process of the aforementioned time-series prediction model includes: constructing time-series training samples using the historical final health scores of power equipment as time series; using a sliding window method, taking the final health scores of the past A time points as input features and the predicted health scores of the next B time points as prediction targets; training the time-series training samples based on an LSTM neural network using the mean squared error between the predicted health scores and the actual health scores of the next B time points as the loss function; and using the trained LSTM neural network, taking the latest final health score sequence of power equipment as input, to predict the predicted health scores of the next B time points.

[0076] In the above embodiments, by inputting the knowledge graph and target weights into the health assessment model, a health time-series curve reflecting the overall operating status of the equipment can be obtained at the current moment. Based on the health time-series curve, predictions can be made to output a health score for future time periods, fully utilizing historical status change trends. Compared to static judgments based solely on health values ​​at a single moment, this method can simultaneously consider the correlation between multiple basic assessment indicators and their dynamic characteristics over time, thereby improving the accuracy and reliability of health status prediction.

[0077] In one embodiment, the method further includes: calculating the slope and segmented slope of the health score of each power device in a future time period; the slope is used to characterize the changing trend of the health score in the future time period; the segmented slope is used to characterize the local changing trend of the health score in each sub-time period; and comparing the slope and segmented slope with preset thresholds respectively to obtain the future change results.

[0078] To more accurately reflect changes in the future health status of power equipment, this embodiment calculates the slope and segmented slope of the predicted future health score. The slope is used to characterize the overall downward or upward trend of the health score over a future period, while the segmented slope is used to characterize the local rate of change of the health score in multiple sub-time periods, in order to identify whether there is a sudden deterioration or accelerated degradation during the change process.

[0079] Optionally, calculating the slope includes: performing a linear fit on the health score at each moment within the future time period, and using the slope of the fitted line as the overall slope; calculating the segmented slope includes: dividing the future time period into multiple sub-time periods of preset length, and performing a linear fit on the health score within each sub-time period, and using the slope of the fitted line as the segmented slope of each sub-time period.

[0080] Then, based on the slope and the comparison results of the segmented slopes with the corresponding thresholds, the future changes are obtained. For example, when the overall slope is lower than the upper threshold, a basic degradation risk is determined; when the overall slope is lower than the safety threshold, and the safety threshold is less than the upper threshold, a severe degradation risk is determined. For example, when the segmented slopes in multiple consecutive sub-time periods are all negative, and the absolute value of the segmented slope in the later sub-time period is greater than the absolute value of the segmented slope in the previous sub-time period, an accelerated degradation risk is determined. Further, if the degradation conditions are met in multiple consecutive sub-time periods, the power equipment is determined to be in a state of continuous degradation.

[0081] In the above embodiments, by calculating the overall slope and the segmented slope respectively and comparing them with preset thresholds, it is possible not only to identify the overall downward trend of the health score in the future, but also to identify rapid changes in local time periods. This allows for the identification of accelerated deterioration risks while identifying basic deterioration risks, making the assessment results of future health status more accurate.

[0082] Furthermore, the aforementioned thresholds include an upper limit threshold and a safety threshold; the upper limit threshold is determined based on the statistical upper limit of the historical health score distribution of each power device under normal operating conditions; the safety threshold is a threshold determined based on the safety margin of the equipment.

[0083] The determination of the statistical upper limit includes: collecting monitoring data generated by power equipment in a healthy operating state without any fault records during a period of operation to form a healthy sample dataset; calculating the 95th percentile of the healthy sample dataset using the percentile method, and using the 95th percentile as the upper limit threshold.

[0084] The determination of the safety threshold includes: collecting historical operation datasets containing multiple identical power equipment, with each record including basic evaluation index values ​​and their corresponding equipment operation status labels (normal or fault); using different values ​​of the index as candidate thresholds, calculating the corresponding true positive rate and false positive rate, and plotting an ROC curve with the true positive rate on the vertical axis and the false positive rate on the horizontal axis; calculating the geometric distance between each candidate point on the ROC curve and the coordinate point (0,1), selecting the point with the smallest distance as the optimal operating point, and the corresponding index value is the safety threshold.

[0085] The true positive rate refers to the proportion of data samples whose indicator values ​​exceed the candidate critical point in all historical data of all devices that eventually fail; the false positive rate refers to the proportion of data samples whose indicator values ​​exceed the candidate critical point in all historical data of all devices that eventually remain in normal operation.

[0086] Furthermore, by extracting indicator values ​​from all "normal" state data, their percentile P1 (e.g., the 80th percentile) and absolute safety limit can be calculated. and in the interval Inner fixed step size (Where N is the preset number of sampling points) A series of candidate critical points are generated, and the above ROC curve calculation is performed on each candidate point to obtain the candidate critical value under different performance.

[0087] In one embodiment, the process of adjusting the target weight includes: inputting the target weight of the previous moment and the corresponding health scores of each basic evaluation indicator as the current state into the weight adjustment model; generating the corresponding weight adjustment amount based on the weight adjustment model; and updating the target weight of the previous moment according to the weight adjustment amount to obtain the adjusted target weight.

[0088] In one embodiment, the process of adjusting the target weight includes: inputting the target weight of the previous moment and the corresponding health scores of each basic evaluation indicator as the current comprehensive state vector into the weight adjustment model; outputting the corresponding weight increment vector based on the weight adjustment model; and updating the target weight of the previous moment according to the weight increment vector to obtain the adjusted target weight.

[0089] The weight adjustment model is built upon a hierarchical reinforcement learning framework. The model includes a policy network (Actor) that outputs the weight increment vector and a value network (Critic) that evaluates the long-term returns of the policy. The policy network takes the current overall state vector as input and outputs the corresponding indicator weight increment vector; the value network takes the same state vector as input and outputs a scalar value representing the long-term returns, used to guide the policy update direction.

[0090] Optionally, the policy network uses the real-time health scores of the previous time-to-date baseline evaluation metrics, the real-time comprehensive health scores of each common health dimension, and the current weight set of all metrics and dimensions as inputs to construct a comprehensive state vector, which is the input feature of the model. In this embodiment, the weight increment vector includes the metric weight increment vector and the dimension weight increment vector.

[0091] The weight adjustment model is trained using a hierarchical reinforcement learning model based on the Actor-Critic framework. The policy network is based on a feedforward neural network, and the input layer dimension is determined by the total number of basic evaluation metrics. Total number of common health dimensions Confirmed, specifically The policy network extracts features sequentially through two fully connected layers (output dimensions of 128 and 64 respectively, using the ReLU activation function), and outputs indicator weight increment vectors and dimension weight increment vectors through a bifurcated output layer. The first fully connected output layer takes the features of that layer as input and outputs a vector with a length equal to the total number of all basic evaluation indicators, directly generating the indicator weight increment. The second fully connected output layer takes the same high-level features as input and outputs a vector with a length equal to the total number of all common health dimensions, directly generating the dimension weight increment. The parameter weight matrix obtained through training can parse decision information from different levels within the same comprehensive state and perform corresponding mappings, enabling the policy network to output hierarchically independent and semantically clear weight increment results.

[0092] After obtaining the increments of the indicator weights and dimension weights, they are merged with the target weights corresponding to the previous time step, and the updated weights are normalized using the Softmax function to obtain the new target weights, specifically: ,in, Wnew represents the weight increment of the policy network output, and Wnew represents the adjusted target weight.

[0093] In this embodiment, to guide the direction of weight adjustment, the value network outputs a scalar value representing long-term returns to the comprehensive state vector, and its training is based on the dominant actor-commentator algorithm. The reward mechanism includes: if the device malfunctions within T1 days (e.g., T1=15) after the agent predicts the health status as attentive or abnormal, a reward is given. If the device is predicted to be healthy but fails within day T2 (e.g., T2=7), then a penalty will be imposed. If abnormal predictions persist for more than T3 days (e.g., T3=14) while the equipment remains normal, a penalty will be applied daily. If the evaluation results are consistent with the inspection / test records, a reward will be given. Training uses the Adam optimizer, whose key hyperparameters can be set as follows: learning rate 0.001, discount factor 0.95, batch size 32, and policy update cycle 24 hours.

[0094] After training, the model is connected to monitoring time series data with a unified timestamp, and interaction data is collected at fixed intervals (e.g., every 24 hours) to incrementally update the network parameters based on historical samples.

[0095] In the above embodiments, by adjusting the target weights, the weights no longer depend on static experience configurations but can be adaptively updated as the device status changes, thereby improving the accuracy and real-time performance of health prediction.

[0096] In an exemplary embodiment, common health dimensions of the power equipment are first determined based on their physical structure and operating principles. Each dimension includes at least one basic evaluation indicator. For example, for transformer equipment, insulation performance, winding temperature rise, oil characteristics, mechanical structure stability, and electromagnetic performance can be used as common health dimensions. Insulation performance can be further subdivided into basic evaluation indicators such as leakage current, dielectric loss value, and partial discharge amplitude. For circuit breaker equipment, contact wear, opening and closing time, mechanical travel, and arc energy can be used as common health dimensions. Opening and closing time can be further subdivided into basic evaluation indicators such as opening coil current waveform and closing spring displacement. These common health dimensions extract electrical, mechanical, thermal, and chemical characteristics from equipment design parameters, operating mechanisms, and fault evolution mechanisms, thereby forming a unified health evaluation dimension system across different equipment types.

[0097] Specifically, an adaptive health function is preset for each basic evaluation indicator, and an update mechanism is set up to be automatically triggered at fixed time intervals. This update mechanism collects operating data from the past period within a preset cycle (e.g., 24 hours or 7 days), and re-executes kernel density estimation to update the adaptive health function of the indicator. This allows the health function to be automatically adjusted as the equipment's operating status changes, thereby solving the problem that fixed membership functions cannot adapt to long-term aging and seasonal operating condition changes.

[0098] After determining the basic assessment indicators, adaptive health function, and common health dimensions, an initial health configuration table is constructed for each power device, categorized by common health dimensions, basic assessment indicators, initial weights, and health scores. Each configuration table is then assigned a version number. By integrating the initial health configuration tables of all devices, a device health configuration library is built for dynamic calling and update management during subsequent operation.

[0099] Subsequently, during equipment operation, real-time power data of each power device in the target power grid system is acquired, and real-time health scores corresponding to each basic evaluation indicator are calculated based on a preset adaptive health function. A knowledge graph is constructed based on the real-time health scores and the target weights corresponding to the devices, and a health time-series curve is generated using the knowledge graph. Then, the health score for future time periods is predicted based on the health time-series curve. Furthermore, the slope and segmented slope of future health scores are calculated based on operational trends and compared with preset thresholds to identify potential accelerated deterioration, stabilization, or recovery trends. Finally, based on the prediction results and operational status, the aforementioned weight adjustment model is invoked to update the target weights, and the updated weights are written into the equipment health configuration library, enabling subsequent health assessments to be performed based on the latest weights.

[0100] In another embodiment, the initial weights of each dimension and indicator can be calculated using the Analytic Hierarchy Process (AHP). First, a three-tiered hierarchical model is constructed, consisting of health assessment objectives, common health dimensions, and basic assessment indicators. Then, pairwise comparisons are performed on elements within the same tier. A judgment matrix is ​​constructed using a 1–9 scaling method, and eigenvalue decomposition is performed on the judgment matrix. The eigenvector corresponding to the largest eigenvalue is used as the weight vector for that tier. Simultaneously, a consistency check is performed on the judgment matrix. If the consistency ratio (CR) is less than 0.10, the weight allocation is considered reasonable; otherwise, the scores of the judgment matrix are adjusted until the standard is met. After obtaining the weight vectors for each level, the initial weights for each common health dimension and each basic assessment indicator are obtained through a weight synthesis method, thus forming the initial weight system for equipment health evaluation.

[0101] In the initial deployment phase, the initial weights are used first. As the power equipment continues to operate and generates new real-time monitoring data, the system dynamically updates the initial weights based on the latest operating status each time it completes a health calculation and generates a health score, thereby obtaining the target weights for the current moment.

[0102] In the above embodiments, the target weight can automatically track changes in device status, avoid evaluation bias caused by the drift of indicator importance over time, and significantly improve the accuracy of health scores.

[0103] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0104] Based on the same inventive concept, this application also provides a health assessment device for power equipment to implement the aforementioned health assessment method for power equipment. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more embodiments of the power equipment health assessment device provided below can be found in the limitations of the power equipment health assessment method described above, and will not be repeated here.

[0105] In an exemplary embodiment, as shown in Figure 2, a health assessment device for power equipment is provided, comprising: an acquisition module, a health calculation module, and a health prediction module, wherein:

[0106] The acquisition module is used to acquire real-time power data of each power device in the target power grid system.

[0107] The health calculation module is used to calculate the health score of each real-time power data according to the pre-calculated health function; the health function is calculated based on the health sample data; the basic evaluation indicators are determined based on the physical structure and working principle of each power device.

[0108] The health prediction module is used to construct a knowledge graph based on health scores, and based on the knowledge graph and the target weights corresponding to each power device, predict the health score of each power device in the future time period; the target weights are obtained by adjusting the target weights corresponding to the previous time.

[0109] In one embodiment, the knowledge graph includes nodes and edges between nodes; nodes are basic evaluation metrics; edges are determined based on the correlation between the basic evaluation metrics; the node features are health scores; the health prediction module includes:

[0110] The health curve calculation unit is used to input the knowledge graph and target weights into the health assessment model to obtain the health time series curve corresponding to the current moment.

[0111] The prediction unit is used to make predictions based on the health time series curve to obtain the health score of each power device in the future time period.

[0112] In one embodiment, the above-described apparatus further includes a trend analysis module, which includes:

[0113] The slope calculation unit is used to calculate the slope and segmented slope of the health score of each power device in the future time period; the slope is used to characterize the changing trend of the health score in the future time period; the segmented slope is used to characterize the local changing trend of the health score in each sub-time period.

[0114] The comparison unit is used to compare the slope and the segmented slope with a preset threshold to obtain the future change result.

[0115] In one embodiment, the threshold includes an upper limit threshold and a safety threshold; the upper limit threshold is determined based on the statistical upper limit of the historical health score distribution of each power device under normal operating conditions; the safety threshold is a threshold determined based on the safety margin of the equipment.

[0116] In one embodiment, the above-mentioned health calculation module includes:

[0117] The kernel density calculation unit is used to estimate the kernel density of healthy sample data to obtain the target bandwidth.

[0118] The probability calculation unit is used to calculate the probability density under different values ​​based on the target bandwidth and the Gaussian kernel function.

[0119] The normalization unit is used to normalize the probability density to obtain the health function.

[0120] In one embodiment, the above-mentioned health prediction module includes:

[0121] The input unit is used to adjust the model by taking the target weights from the previous time step and the corresponding health scores of each basic evaluation indicator as the input weights for the current state.

[0122] The incremental generation unit is used to generate the corresponding weight adjustment amount based on the weight adjustment model.

[0123] The adjustment unit is used to update the target weight of the previous time step according to the weight adjustment amount, so as to obtain the adjusted target weight.

[0124] Each module in the aforementioned health assessment device for power equipment can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0125] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores real-time power data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a method for health assessment of power equipment.

[0126] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0127] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method in any of the above embodiments.

[0128] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method in any of the above embodiments.

[0129] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method in any of the above embodiments.

[0130] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0131] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0132] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for health assessment of electrical equipment, characterized in that, The method includes: Acquire real-time power data of each power device in the target power grid system; the real-time power data includes basic evaluation indicators. The health score of each basic assessment indicator is calculated based on a pre-calculated health score function; the health score function is calculated based on the health sample data. A knowledge graph is constructed based on the health score, and the health score of each power device is predicted in the future time period based on the knowledge graph and the target weights corresponding to each power device; the target weights are obtained by adjusting the target weights corresponding to the previous time.

2. The method according to claim 1, characterized in that, The knowledge graph includes nodes and edges between the nodes; the nodes are basic evaluation indicators; the edges are determined based on the correlation between the basic evaluation indicators; the node feature is a health score. The prediction of the health score of each power device within a future time period based on the knowledge graph and the target weights corresponding to each power device includes: The knowledge graph and the target weights are input into the health assessment model to obtain the health time series curve corresponding to the current moment; Based on the health time series curve, a health score for each of the power devices is obtained in the future time period.

3. The method according to claim 1 or 2, characterized in that, The method further includes: The slope and segmented slope of the health score calculation for each of the aforementioned power devices in the future time period; the slope is used to characterize the changing trend of the health score in the future time period; the segmented slope is used to characterize the local changing trend of the health score in each sub-time period. The slope and the segmented slope are compared with preset thresholds to obtain the future change results.

4. The method according to claim 3, characterized in that, The threshold includes an upper limit threshold and a safety threshold; the upper limit threshold is determined based on the statistical upper limit of the historical health score distribution of each power device under normal operating conditions; the safety threshold is a threshold determined based on the safety margin of the equipment.

5. The method according to claim 1, characterized in that, The method for calculating the health function includes: The target bandwidth is obtained by performing kernel density estimation on the healthy sample data; Based on the target bandwidth and the Gaussian kernel function, the probability density under different values ​​is obtained; The probability density is normalized to obtain the health function.

6. The method according to claim 1, characterized in that, The process of adjusting the target weight includes: The target weights from the previous time step and the corresponding health scores of each of the basic evaluation indicators are used as inputs to the current state to adjust the weights of the model. Generate the corresponding weight adjustment amount based on the weight adjustment model; The target weight at the previous time step is updated based on the weight adjustment amount to obtain the adjusted target weight.

7. A health assessment device for electrical equipment, characterized in that, The device includes: The acquisition module is used to acquire real-time power data of each power device in the target power grid system; the real-time power data includes basic evaluation indicators. The health score calculation module is used to calculate the health score of each of the basic assessment indicators according to a pre-calculated health score function; the health score function is calculated based on the health sample data. The health prediction module is used to construct a knowledge graph based on the health score, and predict the health score of each power device in a future time period based on the knowledge graph and the target weights corresponding to each power device; the target weights are obtained by adjusting the target weights corresponding to the previous time.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images