Energy unit health state evaluation system and method and medium

Abstract

The invention provides an energy unit health state assessment system and method and a medium, relates to the field of energy unit health assessment, and solves the technical problem that early or progressive health state abnormity cannot be effectively identified and early warned in the prior art. The method comprises the following steps: acquiring historical operation data of target equipment in an energy unit during a normal operation period, and constructing a multi-dimensional operation parameter space of the target equipment; grid division is carried out on the multi-dimensional operation parameter space, typical operation state sample points are extracted, clustering is carried out on the typical operation state sample points, and at least one habitual operation area is obtained; constructing an equipment state evidence library based on the typical operation state samples corresponding to the habitual operation areas; based on the equipment state evidence library, analyzing reliability distribution of the real-time operation state of the target equipment relative to a known normal state; and judging the health state of the energy unit based on the reliability distribution. The method is used in the energy unit health assessment process.

Description

Technical Field

[0001] This application relates to the field of energy unit health assessment, and more particularly to an energy unit health status assessment system, method and medium. Background Technology

[0002] With the increasing proportion of new energy power generation (such as photovoltaic and wind power) and energy storage systems in the energy structure, ensuring the reliable operation of key equipment (such as inverters, converters, and battery management systems) has become a core requirement of intelligent operation and maintenance. Key equipment in energy units often experience performance degradation or hidden anomalies due to aging, environmental changes, or control deviations during long-term operation. These problems often lack obvious fault characteristics and are difficult to effectively identify through traditional threshold alarms or supervisory models relying on fault samples, leading to delayed warnings and frequent false alarms, failing to meet the actual needs of high reliability and intelligent operation and maintenance of energy units. Therefore, this application provides an energy unit health status assessment system, method, and medium. Summary of the Invention

[0003] This application provides an energy unit health status assessment system, method, and medium, which solves the technical problem that existing technologies cannot effectively identify and warn of early or progressive health status abnormalities.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for assessing the health status of an energy unit is provided, including: A method for assessing the health status of an energy unit, characterized by comprising the following steps: Acquire historical operating data of the target equipment in the energy unit during normal operation, wherein the historical operating data includes a set of operating parameter observations at multiple time points; Based on the historical operating data, a multi-dimensional operating parameter space for the target device is constructed; the multi-dimensional operating parameter space is divided into grids and typical operating state sample points are extracted; each dimension of the multi-dimensional operating parameters corresponds to one operating parameter. The typical operating state sample points are clustered based on the density peak clustering (DPC) algorithm to obtain at least one habitual operating region; and a device state evidence library is constructed based on the typical operating state samples corresponding to each habitual operating region. Based on the aforementioned equipment status evidence library, an improved evidence K-nearest neighbor classifier with distance suppression classification characteristics is used to assess the confidence distribution of the real-time operating status of the target equipment relative to known normal states, including a first set of confidence values ​​and a second set of confidence values. The first set of confidence values ​​represents the confidence that the current operating status belongs to each known normal state; the second set of confidence values ​​represents the degree to which the current operating status deviates from all known normal states. Based on the first set of confidence values ​​and the second set of confidence values, it is determined whether there is a potential anomaly in the energy unit; if yes, a health status warning signal is sent; otherwise, no action is taken.

[0005] Based on the above technical solutions, in the energy unit health status assessment method provided in this application, key equipment in the energy unit often experiences performance degradation or hidden anomalies due to factors such as aging, environmental changes, or control deviations during long-term operation. These problems often lack obvious fault characteristics and are difficult to effectively identify through traditional threshold alarms or supervised models relying on fault samples. Furthermore, the normal operating conditions of equipment are diverse, and insufficient modeling can easily lead to false alarms or missed alarms. To address these issues, this application proposes an unsupervised health assessment method that relies solely on normal historical operating data: a parameter space is constructed based on multi-dimensional operating parameters; representative typical samples are obtained through grid partitioning and density-aware sampling; density peak clustering (DPC) is then used to automatically identify several habitual operating areas, and an equipment status evidence library is constructed based on these areas; an improved Evidence K-Nearest Neighbor (EKNN) classifier with distance inhibition characteristics is used to calculate the first confidence value (attribution confidence) of real-time data for each known normal state and the second confidence value (deviation degree) for unknown states. By analyzing the dynamic changes of these two types of confidence, risks such as potential anomalies, operating characteristic deviations, or slow degradation can be effectively identified. This solution requires no fault labels and does not rely on manual rules. It features strong adaptability, good interpretability, and excellent anti-interference capabilities, significantly improving the accuracy and foresight of health status early warning.

[0006] In conjunction with the first aspect above, in one possible implementation, constructing the multi-dimensional operating parameter space of the target device based on the historical operating data includes: Several key operating parameters for characterizing the operating status of the target equipment are obtained from the historical operating data; the corresponding value ranges are determined according to the minimum and maximum values ​​of each key operating parameter in the historical operating data, resulting in several value ranges, and these value ranges are set as coordinate axes to construct a multi-dimensional operating parameter space.

[0007] In conjunction with the first aspect above, in one possible implementation, the method for extracting the typical operating state sample points includes: The multi-dimensional parameter operating space is divided into several grid cells; Calculate the data density of grid cell i For grid cell i, based on the corresponding data density Calculate sampling probability ; Calculate the number of samples to be drawn from grid cell i based on the sampling probability. ; ; in, Initialize the adjustment coefficients. , The number of historical data points falling into grid cell j is ceil, where ceil is the rounding up of the real number. The total expected sample size is obtained by summing the expected sample sizes of all grid cells. Where G is the number of grid cells; like Then set the adjustment coefficient. Repeat the extraction process until Grid cell i is randomly drawn without replacement from the historical running data points contained therein. A number of sample points are labeled as typical operating state sample points; where e is the convergence step size control factor.

[0008] In conjunction with the first aspect above, in one possible implementation, dividing the multidimensional parameter operating space into several grid cells includes: Based on the range of each operating parameter in historical operating data [ ; Based on the preset number of segments K, the space of operating parameters in each dimension is divided into K subspaces at equal intervals, with a subspace width of K. = ; By performing a Cartesian product operation on the subspace of each dimension of the operating parameters, multiple non-overlapping grid cells are obtained.

[0009] In conjunction with the first aspect above, in one possible implementation, the construction of a device state evidence library containing state evidence corresponding to each habitual operating region includes: Calculate the local density of sample points under typical operating conditions and relative distance : ; Where j and k are the serial numbers of typical operating state sample points. The Euclidean distance between sample points j and k in a typical operating state is given. This is the preset cutoff distance; ; by x-axis A decision map is constructed using the vertical axis. Points with local density and relative distance both higher than the corresponding preset thresholds are selected from the decision map as cluster centers. The remaining sample points are assigned to the cluster centers to which the nearest maximum local density belongs, forming multiple clusters. Each cluster corresponds to a habitual operating area, and typical operating state sample points are extracted from each habitual operating area to construct an equipment status evidence library.

[0010] In conjunction with the first aspect above, in one possible implementation, the method for obtaining the first set of confidence values ​​includes: M typical sample points with a Euclidean distance of less than a preset distance threshold from the device status evidence library are selected to form a nearest neighbor set, and a habitual operating area label is set for each sample point in the nearest neighbor set. Based on each sample point in the nearest neighbor set With real-time running data The distance calculation uses the weights for distance suppression, including: ; Where r is the index of the M typical operating state sample points whose Euclidean distance is less than a preset distance threshold. To determine the Euclidean distance between the real-time running data and the sample points, The preset distance decay index is used to control the degree of influence of distance on the weights. To prevent division by zero of constants; For each habitual operating region, the weights of all corresponding sample points in the nearest neighbor set are summed to obtain the total weight of the region. Then, the total weights of each region are normalized to obtain the first set of reliability values. ;in, For the r-th habitual operating region, This indicates the confidence level that the current real-time running data belongs to the r-th habitual running region; Based on the first set of reliability values, using the formula The second reliability value was calculated. .

[0011] In conjunction with the first aspect above, in one possible implementation, determining whether the energy unit has a potential anomaly includes: If the second confidence value shows a continuous upward trend over multiple consecutive sampling periods and exceeds the preset warning threshold, it is determined that the energy unit has a potential anomaly. If the normal state corresponding to the maximum value in the first confidence value set of the current period is inconsistent with the historical dominant operating state, and the second confidence value exceeds the preset warning threshold for multiple consecutive periods, then it is determined that the energy unit has an operating characteristic deviation. If at least one of the first confidence values ​​corresponding to a normal state shows a monotonically decreasing trend over multiple consecutive sampling periods, then the energy unit is determined to be slowly degrading. If any of the above conditions are met, a health status warning signal will be sent; otherwise, the current monitoring status will be maintained and no action will be taken.

[0012] In conjunction with the first aspect above, in one possible implementation, acquiring several key operating parameters used to characterize the operating state of the target device includes: Acquire historical operating data of the target device during normal operation, wherein the historical operating data contains time series of N candidate operating parameters; and perform standardization processing on the candidate operating parameters. Calculate the Pearson correlation coefficient between any two candidate running parameters. Construct N The correlation coefficient matrix of N; Traverse the correlation coefficient matrix; if there exist candidate running parameter pairs that satisfy... If the similarity threshold is greater than the preset threshold, the two are determined to be redundant, and one candidate operating parameter is removed. The remaining non-redundant candidate operating parameters are marked as critical operating parameters.

[0013] Secondly, this application provides an energy unit health status assessment, including: an acquisition module, an analysis module, and an early warning module; wherein, the acquisition module is used to acquire historical operating data of a target device in the energy unit during normal operation; the analysis module is used to construct a multi-dimensional operating parameter space of the target device based on the historical operating data; divide the multi-dimensional operating parameter space into grids and extract typical operating state sample points; cluster the typical operating state sample points based on the density peak clustering (DPC) algorithm to obtain at least one habitual operating region; construct an equipment status evidence library based on the typical operating state samples corresponding to each habitual operating region; based on the equipment status evidence library, use an improved evidence K-nearest neighbor classifier with distance suppression classification characteristics to determine the confidence distribution of the real-time operating status of the target device relative to a known normal state; the early warning module is used to determine the health status of the energy unit based on the confidence distribution.

[0014] Thirdly, this application provides a computer-readable storage medium storing instructions that, when executed on an energy unit health status assessment device, cause the energy unit health status assessment device to perform the methods described in the first aspect and any possible implementation thereof.

[0015] This application provides an energy unit health status assessment system, method, and medium that can accurately identify and proactively warn of early equipment anomalies, operational characteristic deviations, and slow degradation without the need for fault samples and manual rules. The scheme first constructs a multi-dimensional operating parameter space based on normal historical operating data and extracts typical operating state samples through a density-aware dynamic sampling strategy, effectively overcoming the problems of high-dimensional redundancy and uneven distribution of the original data. Then, density peak clustering (DPC) is used to automatically discover the equipment's habitual operating areas, constructing an equipment status evidence library that truly reflects its diverse normal operating conditions. Based on this, an improved evidence K-nearest neighbor (EKNN) classifier with distance inhibition characteristics is designed, outputting a first set of confidence values ​​(representing the degree of support for each known normal state) and a second confidence value (representing the uncertainty for unknown or abnormal states). Multi-period trend analysis is used to achieve differentiated discrimination of three types of health risks. This method has the advantages of being completely unsupervised, highly adaptive, highly interpretable, and computationally efficient. It not only avoids the rigidity of traditional threshold alarms and the dependence of supervised learning on labeled data, but also captures early signals of performance degradation before a fault occurs, significantly improving the intelligence level and reliability of energy system operation and maintenance. It is applicable to various energy scenarios such as photovoltaic, energy storage, and wind power.

[0016] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0017] Figure 1 A system architecture diagram of an energy unit health status assessment system provided in this application embodiment; Figure 2 A flowchart illustrating a method for assessing the health status of an energy unit, provided in an embodiment of this application; Figure 3 A flowchart illustrating an analytical method for estimating the total number of samples to be extracted, provided in an embodiment of this application; Figure 4This is a flowchart illustrating a method for analyzing the health status of an energy unit, as provided in an embodiment of this application. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The energy unit health status assessment method provided in this application embodiment can be applied to an energy unit health status assessment system, such as... Figure 1 As shown, the system includes: an acquisition module, an analysis module, and an early warning module; The acquisition module is used to acquire historical operating data of the target equipment in the energy unit during normal operation; The analysis module is used to construct a multi-dimensional operating parameter space for the target equipment based on historical operating data; divide the multi-dimensional operating parameter space into grids and extract typical operating state sample points; cluster the typical operating state sample points based on the density peak clustering (DPC) algorithm to obtain at least one habitual operating region; construct an equipment state evidence library based on the typical operating state samples corresponding to each habitual operating region; and use an improved evidence K-nearest neighbor classifier with distance suppression classification characteristics to determine the confidence distribution of the real-time operating state of the target equipment relative to the known normal state based on the equipment state evidence library. The early warning module is used to determine whether there are potential anomalies in the energy unit based on the confidence distribution; if so, a health status early warning signal is sent; otherwise, no action is taken.

[0020] To address the technical problem that existing technologies cannot effectively identify and warn of early or progressive health status anomalies, this application provides a method for assessing the health status of an energy unit. The method includes: acquiring historical operating data of a target device in the energy unit during normal operation, the historical operating data containing a set of observed operating parameters at multiple time points; constructing a multi-dimensional operating parameter space for the target device based on the historical operating data; dividing the multi-dimensional operating parameter space into a grid and extracting typical operating state sample points; each dimension of the multi-dimensional operating parameters corresponds to one operating parameter; clustering the typical operating state sample points based on the density peak clustering (DPC) algorithm to obtain at least one habitual operating region; constructing an equipment status evidence library based on the typical operating state samples corresponding to each habitual operating region; and, based on the equipment status evidence library, using an improved K-nearest neighbor classifier with distance suppression classification characteristics to assess the confidence distribution of the real-time operating status of the target device relative to a known normal state, including a first confidence value. The system comprises two sets of reliability values: a first set of reliability values ​​representing the confidence level of the current operating state belonging to each known normal state, and a second set of reliability values ​​representing the degree to which the current operating state deviates from all known normal states. Based on the first set of reliability values ​​and the second set of reliability values, the system determines whether there are potential anomalies in the energy unit. If yes, a health status warning signal is sent; otherwise, no action is taken. Based on this, the application divides historical data into grids based on a multi-dimensional operating parameter space and extracts typical samples using a density-aware strategy to effectively eliminate redundancy and retain the diversity of operating conditions. Density peak clustering (DPC) is used to automatically identify the habitual operating areas of the equipment, forming a structured equipment status evidence library that accurately depicts its normal behavior patterns. An improved evidence K-nearest neighbor (EKNN) classifier with distance inhibition characteristics is used to output a first set of reliability values ​​(reflecting the confidence level of belonging to each known normal state) and a second set of reliability values ​​(quantifying the degree of deviation from unknown states), achieving a refined characterization of the current operating state. By analyzing the dynamic changes of the two types of reliability, risks such as potential anomalies, characteristic shifts, or slow degradation can be accurately identified. This solution requires no fault samples and does not rely on manual thresholds. It is highly robust, interpretable, and has good engineering applicability, significantly improving early warning capabilities and the level of intelligent operation and maintenance.

[0021] like Figure 2 As shown in the embodiment of this application, a method for assessing the health status of an energy unit includes: S201. Obtain historical operating data of the target equipment in the energy unit during normal operation.

[0022] The historical operational data includes a set of operational parameter observations at multiple points in time.

[0023] S202. Based on historical operating data, construct a multi-dimensional operating parameter space for the target equipment; divide the multi-dimensional operating parameter space into grids and extract typical operating state sample points.

[0024] In the multidimensional operating parameters, each dimension corresponds to one operating parameter.

[0025] S203. Cluster typical operating state sample points based on the density peak clustering (DPC) algorithm to obtain at least one habitual operating region; construct an equipment status evidence library based on the typical operating state samples corresponding to each habitual operating region.

[0026] S204. Based on the equipment status evidence library, an improved evidence K-nearest neighbor classifier with distance inhibition classification characteristics is used to determine the confidence distribution of the real-time operating status of the target equipment relative to the known normal status, including a first confidence value set and a second confidence value.

[0027] The first set of reliability values ​​represents the confidence level of the current operating state belonging to each of the known normal states; the second reliability value represents the degree to which the current operating state deviates from all known normal states.

[0028] Based on the first set of confidence values ​​and the second set of confidence values, it is determined whether there is a potential anomaly in the energy unit; if yes, a health status warning signal is sent; otherwise, no action is taken.

[0029] Based on the above technical solutions, in the operation and maintenance of energy systems, target equipment (such as inverters and energy storage converters) is susceptible to aging, environmental disturbances, or control deviations during long-term operation, leading to slow performance degradation or sudden anomalies. Traditional fault diagnosis methods based on fixed thresholds or supervised learning are difficult to effectively identify early health degradation states that are unlabeled, without mutations, but have evolutionary characteristics, and lack sufficient modeling of the diversity of normal operating conditions, easily causing missed or false alarms. To address this, this technical solution proposes an unsupervised health assessment framework driven entirely by normal historical data: First, a multi-dimensional operating parameter space is constructed and typical samples are intelligently extracted to avoid redundancy and skewness; then, density peak clustering (DPC) is used to automatically discover the habitual operating areas of the equipment, forming a structured equipment state evidence library; furthermore, an improved evidence K-nearest neighbor (EKNN) classifier with distance inhibition characteristics is designed to output a first confidence value for each known normal state and a second confidence value for unknown states. This dual-confidence mechanism can accurately quantify the "attribution" and "abnormality" of the current state, supporting multi-dimensional discrimination of potential anomalies, characteristic deviations, and slow degradation. This solution requires no fault data and does not rely on manual rules. It has advantages such as adaptability, interpretability, and strong noise resistance, which significantly improves the foresight and reliability of early warning of the health status of energy equipment.

[0030] In one possible implementation of the embodiments of this application, such as Figure 3As shown, the above S202 can be specifically implemented through the following S301, S302, S303 and S304, which are explained in detail below: S301. Obtain several key operating parameters from historical operating data to characterize the operating status of the target equipment. The specific method for obtaining these parameters is as follows: S3011. Obtain historical operating data of the target device during normal operation. The historical operating data contains time series of N candidate operating parameters. Standardize the candidate operating parameters. S3012. Calculate the Pearson correlation coefficient between any two candidate operating parameters. Construct N The correlation coefficient matrix of N; S3013. Traverse the correlation coefficient matrix. If there exist candidate running parameter pairs that satisfy... If the similarity threshold is greater than the preset threshold, the two are determined to be redundant, and one candidate operating parameter is removed. The remaining non-redundant candidate operating parameters are marked as critical operating parameters.

[0031] S302. Determine the corresponding value intervals based on the minimum and maximum values ​​of each key operating parameter in the historical operating data, obtain several value intervals, and set the several value intervals as coordinate axes to construct a multi-dimensional operating parameter space.

[0032] S303, Based on the range of each operating parameter in historical operating data [ ; Based on the preset number of segments K, the space of operating parameters in each dimension is divided into K subspaces at equal intervals, with a subspace width of K. = ; By performing a Cartesian product operation on the subspace of each dimension of the operating parameters, multiple non-overlapping grid cells are obtained.

[0033] For example, if there are 2-dimensional parameters A, If each dimension is divided into K=3 segments, then there are a total of 3 3 = 9 grid cells; each grid cell consists of a subspace A multiplied by a subspace B, which is the Cartesian product.

[0034] S304, Calculate the data density of grid cell i ;in, The number of historical running data points that fall into grid cell i. Let i be the volume of grid cell i; For grid cell i, based on the corresponding data density Calculate sampling probability Where b is the adjustment coefficient, b>0, usually taken as 0.1~1, controlling the degree of influence of density on sampling probability; when When it is larger (high density), then The sampling probability is small, avoiding oversampling; when When it is smaller (lower density), then The sampling probability is high, ensuring that rare states are captured.

[0035] S305. Calculate the number of samples to be drawn from grid cell i based on the sampling probability. ; ; in, Initialize the adjustment coefficients. ceil is used to round up a real number.

[0036] It should be pointed out that, Let i be the number of samples to be drawn from grid cell i. This means that the number of data points cannot exceed the actual number of data points in grid cell i.

[0037] The total expected sample size is obtained by summing the expected sample sizes of all grid cells. Where G is the number of grid cells; like Then set the adjustment coefficient. Repeat the extraction process until Grid cell i is randomly drawn without replacement from the historical running data points contained therein. Each sample point is marked as a typical operating state sample point; Where e is the convergence step size control factor, used to control the adjustment coefficient. The magnitude of the change in each iteration; if ,but Increase the expected number of samples to be drawn for each grid cell; if ,but Reduce the expected number of samples to be drawn for each grid cell.

[0038] It should be noted that this scheme adopts the DBS algorithm, which extracts more samples in high-density areas (areas with concentrated normal operating conditions) and fewer samples in low-density areas (boundaries or rare operating conditions).

[0039] Based on the above technical solutions, in the health status assessment of energy equipment, the original historical operating data is usually characterized by high dimensionality, strong redundancy, and uneven distribution. Directly using this data for modeling not only incurs high computational costs but is also susceptible to collinearity interference and struggles to effectively represent rare but critical normal operating conditions, leading to biases in subsequent clustering and state identification. Therefore, a method that can automatically filter key parameters and intelligently extract representative samples is urgently needed. This solution first eliminates redundant parameters using the Pearson correlation coefficient to construct a low-dimensional, highly discriminative multidimensional operating parameter space. Then, it employs a dynamic DBS sampling strategy based on density bias: "high-density downweighting and low-density upweighting" is implemented based on the calculated expected number of samples to ensure that rare states are not overwhelmed. Finally, a closed-loop iterative mechanism is constructed by introducing adjustment coefficients and convergence factors to precisely control the total number of samples to a preset value. This method balances sample representativeness, diversity, and controllable quantity, significantly improving the accuracy and efficiency of subsequent DPC clustering and EKNN classification, and providing a high-quality data foundation for unsupervised health assessment.

[0040] In one possible implementation of this application embodiment, the above-mentioned S203 can be specifically implemented by the following S401 and S402, which are described in detail below: S401. Calculate the local density of sample points under typical operating conditions. and relative distance : ; Where j and k are the serial numbers of typical operating state sample points. The Euclidean distance between sample points j and k in a typical operating state is given. This is the preset cutoff distance.

[0041] It should be pointed out that, The larger the value, the denser the area where the typical operating state sample point j is located, and the greater the possibility that it is a cluster center.

[0042] ; It should be noted that when j is not a point of maximum density, This represents the nearest Euclidean distance from a typical running state sample point j to all points with a higher local density than itself. It means that if a point has no higher density neighbors around it, it is farther away from the "high density area" and is more likely to be a cluster center. If j is the point with the maximum global density, then The goal is to find the maximum Euclidean distance to all other points, ensuring it has a large value in the decision graph. Therefore, it is given priority to be selected as the center.

[0043] Therefore The larger the value, the farther the typical operating state sample point j is from the high-density area. If simultaneously The larger the value, the more likely it is to be a cluster center.

[0044] S402, with x-axis A decision map is constructed using the vertical axis. Points with local density and relative distance both higher than the corresponding preset thresholds are selected from the decision map as cluster centers. The remaining sample points are assigned to the cluster centers to which the nearest maximum local density belongs, forming multiple clusters. Each cluster corresponds to a habitual operating area, and typical operating state sample points are extracted from each habitual operating area to construct an equipment status evidence library.

[0045] Based on the above technical solutions, while historical operating data can reflect normal operating conditions in equipment health status assessment, its complex distribution and mixed patterns make it difficult to effectively distinguish different typical operating modes if directly used for status identification, and it is easily affected by noise or sparse samples. Therefore, there is an urgent need for a method that can automatically discover and structurally characterize the normal operating modes of equipment. This solution adopts the Density Peak Clustering (DPC) algorithm, which constructs a decision map to adaptively identify cluster centers by calculating the local density and relative distance of each typical sample point, without the need to preset the number of clusters. Higher local density indicates that the point is in a dense region, and a larger relative distance indicates that it is far from higher density points. The combination of these two factors can accurately locate the core of each stable operating mode. Subsequently, the samples are assigned to the corresponding centers to form clusters, and each cluster represents a "habitual operating region". Finally, typical samples are extracted from each region to construct an equipment status evidence library, providing a high-quality, structured normal reference template for subsequent status identification based on evidence theory. This method has the advantages of being unsupervised, adaptive, and highly noise-resistant, significantly improving the accuracy and robustness of health assessment.

[0046] In one possible implementation of the embodiments of this application, such as Figure 4 As shown, the above S204 can be specifically implemented through the following S501, S502, S503, S504 and S505, which are explained in detail below: S501. Select M typical operating status sample points from the equipment status evidence library whose Euclidean distance from the real-time operating data is less than a preset distance threshold, form a nearest neighbor set, and set the habitual operating area label to which each sample point in the nearest neighbor set belongs.

[0047] S502, Based on each sample point in the nearest neighbor set With real-time running data The distance calculation uses the weights for distance suppression, including: ; Where r is the index of the M typical operating state sample points whose Euclidean distance is less than a preset distance threshold. To determine the Euclidean distance between the real-time running data and the sample points, The preset distance decay index is used to control the degree of influence of distance on the weights. To prevent division by zero of constants; It should be noted that the smaller the Euclidean distance between the real-time running data and the sample points, the greater the weight.

[0048] S503. For each habitual operating region, sum the weights of all corresponding sample points in its nearest neighbor set to obtain the total weight of the region. Then, normalize the total weights of each region to obtain the first set of reliability values. ;in, For the r-th habitual operating region, This indicates the confidence level that the current real-time running data belongs to the r-th habitual running region.

[0049] S504. Calculate the second reliability value based on the first reliability value set. .

[0050] S505. If the second confidence value shows a continuous upward trend in multiple consecutive sampling periods and exceeds the preset warning threshold, it is determined that there is a potential anomaly in the energy unit. It should be noted that the second confidence value represents the degree to which the current state cannot be explained by any known normal pattern; if it continues to rise in consecutive sampling periods (e.g., 0.12→0.15→0.18), it indicates that the behavior of the target device is deviating more and more from all historical normal operating conditions; this usually indicates early fault budding (e.g., sensor drift, poor contact), new unmodeled disturbance sources (e.g., sudden increase in power grid harmonics) or control logic abnormalities.

[0051] If the normal state corresponding to the maximum value in the first confidence value set of the current period is inconsistent with the historical dominant operating state, and the second confidence value exceeds the preset warning threshold for multiple consecutive periods, then it is determined that the energy unit has an operating characteristic deviation. It should be noted that the target device should operate stably within a certain familiar range under the same external conditions (such as load, environment, and commands); if the dominant state suddenly changes (e.g., from the high-efficiency range), Switch to fluctuation zone If the new state cannot be fully explained, it indicates that the inherent characteristics of the equipment have changed, such as: the inverter MPPT algorithm is inaccurate, the internal resistance of the energy storage battery increases, causing the charge and discharge curve to deviate, or the wind turbine pitch control is deviated.

[0052] If at least one of the first confidence values ​​corresponding to a normal state shows a monotonically decreasing trend over multiple consecutive sampling periods, then the energy unit is determined to be slowly degrading. It should be noted that a certain habitual operating area (such as...) =First confidence value of "Fully loaded and highly efficient operation") If it continues to decrease monotonically (e.g., 0.75→0.60→0.45→0.30), it indicates that even under the same input conditions, the equipment is finding it increasingly difficult to return to its optimal operating point. Typical reasons include: dust accumulation / aging of photovoltaic modules → reduced output power under the same irradiance, decreased efficiency of the heat dissipation system → accelerated temperature rise, forced derating operation or mechanical wear → increased vibration, and the control system actively limiting the speed.

[0053] If any of the above conditions are met, a health status warning signal will be sent; otherwise, the current monitoring status will be maintained and no action will be taken.

[0054] Based on the aforementioned technical solutions, the health status of energy equipment during long-term operation is difficult to effectively monitor using traditional threshold alarms, especially when facing abnormal behaviors such as early failures, performance deviations, or slow degradation that have cumulative harm without obvious sudden changes. Relying on human experience or static models is prone to missed or false alarms and cannot distinguish between normal operating condition switching and actual degradation. This solution proposes an intelligent early warning mechanism based on evidence theory and dynamic nearest neighbor weighting: First, an equipment status evidence library composed of habitual operating areas is constructed; then, the support degree of real-time data for each normal mode is quantified by distance suppression weights to generate a first set of confidence values; and a second confidence value is introduced to characterize the uncertainty of unknown states. On this basis, three types of criteria are designed: potential anomalies, operating characteristic deviations, and slow degradation, to achieve accurate identification of different degradation modes. This solution does not require fault samples and can achieve unsupervised, interpretable, and multi-level health status perception based solely on normal data, significantly improving the foresight and robustness of early warnings and providing a reliable basis for predictive maintenance.

[0055] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

Claims

1. A method for assessing the health status of an energy unit, characterized in that, include: Acquire historical operating data of the target equipment in the energy unit during normal operation, wherein the historical operating data includes a set of operating parameter observations at multiple time points; Based on the historical operating data, a multi-dimensional operating parameter space for the target device is constructed; The multidimensional operating parameter space is divided into multiple grid cells, and typical operating state sample points are extracted. Each dimension in the multidimensional operating parameters corresponds to one operating parameter; The typical operating state sample points are clustered based on the density peak clustering (DPC) algorithm to obtain at least one habitual operating region; and a device state evidence library is constructed based on the typical operating state samples corresponding to each habitual operating region. Based on the aforementioned equipment status evidence library, an improved evidence K-nearest neighbor classifier with distance suppression classification characteristics is used to assess the confidence distribution of the real-time operating status of the target equipment relative to known normal states, including a first set of confidence values ​​and a second set of confidence values. The first set of confidence values ​​represents the confidence that the current operating status belongs to each known normal state; the second set of confidence values ​​represents the degree to which the current operating status deviates from all known normal states. Based on the first set of confidence values ​​and the second set of confidence values, it is determined whether there is a potential anomaly in the energy unit; if yes, a health status warning signal is sent; otherwise, no action is taken.

2. The method for assessing the health status of an energy unit according to claim 1, characterized in that, The construction of a multi-dimensional operating parameter space for the target device based on the historical operating data includes: Several key operating parameters for characterizing the operating status of the target equipment are obtained from the historical operating data; the corresponding value ranges are determined according to the minimum and maximum values ​​of each key operating parameter in the historical operating data, resulting in several value ranges, and these value ranges are set as coordinate axes to construct a multi-dimensional operating parameter space.

3. The method for assessing the health status of an energy unit according to claim 1, characterized in that, The method for extracting typical operating state sample points includes: The multi-dimensional parameter operating space is divided into several grid cells; Calculate the data density of grid cell i For grid cell i, based on the corresponding data density Calculate sampling probability ; Calculate the number of samples to be drawn from grid cell i based on the sampling probability. ; ； in, Initialize the adjustment coefficients. , The number of historical data points falling into grid cell j is ceil, where ceil is the rounding up of the real number. The total expected sample size is obtained by summing the expected sample sizes of all grid cells. Where G is the number of grid cells; like Then set the adjustment coefficient. Repeat the extraction process until Grid cell i is randomly drawn without replacement from the historical running data points contained therein. A number of sample points are labeled as typical operating state sample points; where e is the convergence step size control factor.

4. The method for assessing the health status of an energy unit according to claim 3, characterized in that, The process of dividing the multidimensional parameter operating space into several grid cells includes: Based on the range of each operating parameter in historical operating data [ ; Based on the preset number of segments K, the space of operating parameters in each dimension is divided into K subspaces at equal intervals, with a subspace width of K. = ; By performing a Cartesian product operation on the subspace of each dimension of the operating parameters, multiple non-overlapping grid cells are obtained.

5. The method for assessing the health status of an energy unit according to claim 1, characterized in that, The construction of the device status evidence library, which includes status evidence corresponding to each habitual operating region, includes: Calculate the local density of sample points under typical operating conditions and relative distance : ； Where j and k are the serial numbers of typical operating state sample points, The Euclidean distance between sample points j and k in a typical operating state is given. This is the preset cutoff distance; ； by x-axis A decision map is constructed using the vertical axis. Points with local density and relative distance both higher than the corresponding preset thresholds are selected from the decision map as cluster centers. The remaining sample points are assigned to the cluster centers to which the nearest maximum local density belongs, forming multiple clusters. Each cluster corresponds to a habitual operating area, and typical operating state sample points are extracted from each habitual operating area to construct an equipment status evidence library.

6. The method for assessing the health status of an energy unit according to claim 1, characterized in that, The method for obtaining the first set of confidence values ​​includes: M typical sample points with a Euclidean distance of less than a preset distance threshold from the device status evidence library are selected to form a nearest neighbor set, and a habitual operating area label is set for each sample point in the nearest neighbor set. Based on each sample point in the nearest neighbor set With real-time running data The distance calculation uses the weights for distance suppression, including: ； Where r is the index of the M typical operating state sample points whose Euclidean distance is less than a preset distance threshold. To determine the Euclidean distance between the real-time running data and the sample points, The preset distance decay index is used to control the degree to which distance affects the weight. To prevent division by zero of constants; For each habitual operating region, the weights of all corresponding sample points in the nearest neighbor set are summed to obtain the total weight of the region. Then, the total weights of each region are normalized to obtain the first set of reliability values. ;in, For the r-th habitual operating region, This indicates the confidence level that the current real-time running data belongs to the r-th habitual running region; Based on the first set of confidence scores, using the formula The second reliability value was calculated. .

7. The method for assessing the health status of an energy unit according to claim 1, characterized in that, Determining whether the energy unit has potential anomalies includes: If the second confidence value shows a continuous upward trend over multiple consecutive sampling periods and exceeds the preset warning threshold, it is determined that the energy unit has a potential abnormality and a health status warning signal is sent. If the normal state corresponding to the maximum value in the first confidence value set of the current period is inconsistent with the historical dominant operating state, and the second confidence value exceeds the preset warning threshold for multiple consecutive periods, then it is determined that the energy unit has an operating characteristic deviation and a health status warning signal is sent. If at least one of the first confidence values ​​corresponding to a normal state shows a monotonically decreasing trend over multiple consecutive sampling periods, it is determined that the energy unit is undergoing slow degradation, and a health status warning signal is sent.

8. The method for assessing the health status of an energy unit according to claim 1, characterized in that, The acquisition of several key operating parameters used to characterize the operating status of the target device includes: Acquire historical operating data of the target device during normal operation, wherein the historical operating data contains time series of N candidate operating parameters; and perform standardization processing on the candidate operating parameters. Calculate the Pearson correlation coefficient between any two candidate running parameters. Construct N The correlation coefficient matrix of N; Traverse the correlation coefficient matrix; if there exist candidate running parameter pairs that satisfy... If the similarity threshold is greater than the preset threshold, the two are determined to be redundant, and one candidate operating parameter is removed. The remaining non-redundant candidate operating parameters are marked as critical operating parameters.

9. An energy unit health status assessment system, operating based on the energy unit health status assessment method according to any one of claims 1-8, characterized in that, It includes an acquisition module, an analysis module, and an early warning module; The acquisition module is used to acquire historical operating data of the target equipment in the energy unit during normal operation; The analysis module is used to construct a multi-dimensional operating parameter space for the target device based on the historical operating data. The multidimensional operating parameter space is divided into grids and typical operating state sample points are extracted. The typical operating state sample points are clustered based on the density peak clustering (DPC) algorithm to obtain at least one habitual operating region; and a device state evidence library is constructed based on the typical operating state samples corresponding to each habitual operating region. Based on the aforementioned equipment status evidence library, an improved evidence K-nearest neighbor classifier with distance inhibition classification characteristics is used to determine the confidence distribution of the real-time operating status of the target equipment relative to its known normal state. The early warning module is used to determine whether there is a potential anomaly in the energy unit based on the confidence distribution; If yes, then send a health status warning signal; No, then no action will be taken.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the energy unit health status assessment method according to any one of claims 1-8.

Images