Power station anomaly detection method, device, equipment and medium

Abstract

The invention discloses a power station anomaly detection method. The method comprises the following steps: determining a reference power station and a to-be-detected power station from a distributed photovoltaic power station system; obtaining normalized power daily curves of the to-be-detected power stations and the reference power station, and determining a first candidate low-power power station and a first confidence coefficient from the to-be-detected power stations through clustering according to the normalized power daily curves; meanwhile, the power loss rate of the to-be-detected power stations relative to the reference power station is determined through a normalized power daily curve, and a second candidate low-power power station and a second confidence coefficient are determined from the to-be-detected power stations according to the power loss rate, so that a target low-power power station is determined according to the selected low-power power station and the confidence coefficient; and finally, determining an abnormal evaluation result of the target low-power power station by obtaining the inverter direct-current side input power, the alternating-current three-phase voltage and the inverter safety voltage of the target low-power power station. According to the method, multi-source feature parameters can be effectively fused, and the anomaly detection precision is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power station anomaly detection, and in particular to a power station anomaly detection method, device, equipment and medium. BACKGROUND

[0002] With the global energy structure accelerating towards clean and low-carbon transformation, photovoltaic power generation, as an important part of renewable energy, has a continuously rapid growth in installed capacity and is widely used in various scenarios. Under the background of rapid development of distributed energy system, distributed small photovoltaic systems have become an important force to improve regional energy self-sufficiency and build distributed energy networks due to their advantages such as flexible layout, short construction period, and low investment threshold.

[0003] However, compared with centralized photovoltaic power stations, distributed photovoltaic systems are more susceptible to the following factors during operation: limited installation environment (shading, non-uniform orientation), uneven equipment quality (mostly using low-end inverters or components), lack of professional operation and maintenance (lack of remote monitoring capability), etc. These factors can easily cause abnormal operation of photovoltaic power stations, resulting in reduced power generation efficiency, abnormal power output, and even safety risks. Currently, abnormal diagnosis for low-power states still mainly relies on manual inspection or traditional threshold alarm mechanisms, which have obvious limitations. Therefore, there is an urgent need for a power station anomaly detection method that can integrate multi-source data and dynamically identify power station anomalies and their causes. SUMMARY

[0004] Embodiments of the present application provide a power station anomaly detection method, device, equipment and medium to solve the problem of limited detection accuracy due to single detection method in related technologies.

[0005] In a first aspect, the present disclosure provides a power station anomaly detection method suitable for a distributed photovoltaic power station system, comprising: determining a reference power station and a to-be-detected power station other than the reference power station from the distributed photovoltaic power station system; obtaining normalized power daily curves of the to-be-detected power station and the reference power station, and clustering the to-be-detected power station according to the normalized power daily curves to obtain a plurality of clustering clusters; based on the normalized power daily mean curve of each clustering cluster and the normalized power daily curve of the reference power station, determining a first candidate low-power power station from the to-be-detected power station and a first confidence degree corresponding to the first candidate low-power power station; based on the normalized power daily curves of the to-be-detected power station and the reference power station, determining a power loss rate of the to-be-detected power station compared to the reference power station, and according to the power loss rate, determining a second candidate low-power power station from the to-be-detected power station and a second confidence degree corresponding to the second candidate low-power power station; determine a target low-power power station based on the first candidate low-power power station and the first confidence level, and the second candidate low-power power station and the second confidence level; obtain an inverter DC side input power, an AC three-phase voltage and an inverter safety voltage of the target low-power power station, and determine an abnormality evaluation result of the target low-power power station based on the inverter DC side input power, the AC three-phase voltage and the inverter safety voltage.

[0006] In an embodiment, the to-be-detected power stations are clustered according to the normalized power daily curves, and a plurality of clustering clusters are obtained, including: sampling the normalized power daily curves based on a first preset time granularity, and determining a daily operation feature vector of the to-be-detected power station; clustering the to-be-detected power stations based on the daily operation feature vector, and obtaining a plurality of clustering clusters.

[0007] In an embodiment, based on the normalized power daily mean curve of each clustering cluster and the normalized power daily curve of the reference power station, a first candidate low-power power station is determined from the to-be-detected power stations, and a first confidence level corresponding to the first candidate low-power power station, including: performing mean value processing based on the normalized power daily curve corresponding to each clustering cluster, and determining a normalized power daily mean curve of each clustering cluster; comparing the normalized power daily mean curve with the normalized power daily curve of the reference power station to determine a relative power ratio of each clustering cluster; using a preset relative power ratio threshold and the relative power ratio, the to-be-detected power station corresponding to the clustering cluster with the relative power ratio less than the preset relative power ratio threshold is determined as the first candidate low-power power station, and the relative power ratio corresponding to the first candidate low-power power station is determined as the first confidence level.

[0008] In an embodiment, based on the normalized power daily curves of the to-be-detected power station and the reference power station, a power loss rate of the to-be-detected power station compared with the reference power station is determined, and a second candidate low-power power station is determined from the to-be-detected power stations according to the power loss rate, and a second confidence level corresponding to the second candidate low-power power station, including: sampling the normalized power daily curves based on a second preset time granularity to obtain resampled normalized powers of the to-be-detected power station and the reference power station; determining power loss rates of the to-be-detected power station and the reference power station at a plurality of sampling time points based on the resampled normalized powers; using a preset power loss threshold and the power loss rate comparison, the sampling time point with the power loss rate greater than or equal to the preset power loss threshold is determined as an abnormal time point; The power station to be detected with the number of abnormal time greater than or equal to the preset abnormal number threshold is determined as a second candidate low-power power station, and a ratio of the number of abnormal times of the second candidate low-power power station to the total number of sampling time is determined as a second confidence.

[0009] In an embodiment, the target low-power power station is determined based on the first candidate low-power power station and the first confidence, and the second candidate low-power power station and the second confidence, comprising: The fusion confidence of the first candidate low-power power station and the second candidate low-power power station is determined by weighted summation based on the first confidence and the first preset weight corresponding to the first confidence, and the second confidence and the second preset weight corresponding to the second confidence; The first candidate low-power power station and the second candidate low-power power station with the fusion confidence greater than or equal to the preset confidence threshold are determined as the target low-power power station based on the fusion confidence and the preset confidence threshold.

[0010] In an embodiment, the abnormal evaluation result of the target low-power power station is determined based on the input power of the inverter DC side, the three-phase AC voltage and the inverter safety voltage, comprising: The input power stop time of the target low-power power station is determined based on the input power of the inverter DC side; The critical overvoltage time and the critical peak regulation time of the target low-power power station are determined based on the three-phase AC voltage and the inverter safety voltage; The same time in the input power stop time and the critical overvoltage time is determined as the grid overvoltage time, and the same time in the input power stop time and the critical peak regulation time is determined as the grid peak regulation time; When the continuous time of the grid overvoltage time is greater than or equal to the preset overvoltage duration threshold, it is determined that the abnormal evaluation result is that the target low-power power station has grid overvoltage abnormality; When the continuous time of the grid peak regulation time is greater than or equal to the preset peak regulation duration threshold, it is determined that the abnormal evaluation result is that the target low-power power station has grid peak regulation abnormality.

[0011] In an embodiment, the method further comprises: When the target low-power power station does not have grid overvoltage abnormality and grid peak regulation abnormality, the output current of the multi-channel controller and the string specification of the power station group of the target low-power power station are obtained; The actual current ratio of each channel of the controller is determined based on the output current of the multi-channel controller and the string specification of the power station group, and the time when each actual current ratio is greater than or equal to the preset current ratio threshold is determined as the string abnormal time; The input power peak value and the input power median value of the target low-power power station in the preset sampling interval are determined based on the input power of the inverter DC side to obtain the input power median absolute deviation of the target low-power power station. determining that the abnormality evaluation result is that the target low-power power station has flexible power limiting when the product of the input power peak value and the preset deviation threshold is greater than the median absolute deviation of the input power; determining that the abnormality evaluation result is that the target low-power power station has string abnormality when the number of times of the string abnormality time is greater than or equal to the preset string abnormality threshold.

[0012] In a second aspect, the present disclosure provides a power station abnormality detection device, comprising: a power station classification module configured to determine a reference power station from a distributed photovoltaic power station system, and a to-be-detected power station other than the reference power station; a data acquisition module configured to acquire a normalized power daily curve of the to-be-detected power station and the reference power station, and cluster the to-be-detected power station according to the normalized power daily curve to obtain a plurality of clustering clusters; a first power station identification module configured to determine a first candidate low-power power station from the to-be-detected power station based on a normalized power daily mean curve of each clustering cluster and the normalized power daily curve of the reference power station, and a first confidence degree corresponding to the first candidate low-power power station; a second power station identification module configured to determine a power loss rate of the to-be-detected power station compared with the reference power station based on the normalized power daily curve of the to-be-detected power station and the reference power station, and determine a second candidate low-power power station from the to-be-detected power station according to the power loss rate, and a second confidence degree corresponding to the second candidate low-power power station; a target low-power power station identification module configured to determine a target low-power power station based on the first candidate low-power power station and the first confidence degree, and the second candidate low-power power station and the second confidence degree; an abnormality evaluation module configured to acquire an input power of a direct current side of an inverter of the target low-power power station, an alternating current three-phase voltage, and an inverter safety voltage, and determine an abnormality evaluation result of the target low-power power station based on the input power of the direct current side of the inverter, the alternating current three-phase voltage, and the inverter safety voltage.

[0013] In a third aspect, the present disclosure provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the power station abnormality detection method described above when executing the computer program. In a fourth aspect, the present disclosure provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the power station abnormality detection method described above.

[0014] In the scheme implemented by the power plant anomaly detection method, device, equipment and medium, the reference power plant and the power plant to be detected are determined from the distributed photovoltaic power plant system; the normalized power daily curve of the power plant to be detected and the reference power plant is obtained, and a first candidate low-power power plant and a first confidence level are determined from the power plant to be detected according to the normalized power daily curve through clustering; at the same time, the power loss rate of the power plant to be detected compared with the reference power plant is determined through the normalized power daily curve, and a second candidate low-power power plant and a second confidence level are determined from the power plant to be detected according to the power loss rate, so as to determine the target low-power power plant according to the first candidate low-power power plant, the second candidate low-power power plant, the first confidence level and the second confidence level; finally, the abnormal evaluation result of the target low-power power plant is determined by obtaining the inverter direct current side input power, the alternating current three-phase voltage and the inverter safety voltage of the target low-power power plant. The method of the present disclosure combines dynamic clustering analysis with fixed threshold comparison, and performs abnormal identification on the power plant to be detected from multiple dimensions, thereby improving the accuracy and robustness of low-power abnormality determination; at the same time, the abnormal evaluation result of the target low-power power plant is determined through multi-dimensional parameter analysis on the target low-power power plant, thereby further improving the accuracy of the abnormal evaluation result. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0016] Figure 1 is a flowchart of a power plant anomaly detection method in an embodiment of the present application; Figure 2 is another flowchart of a power plant anomaly detection method in an embodiment of the present application; Figure 3 is another flowchart of a power plant anomaly detection method in an embodiment of the present application; Figure 4 is another flowchart of a power plant anomaly detection method in an embodiment of the present application; Figure 5 is another flowchart of a power plant anomaly detection method in an embodiment of the present application; Figure 6 is another flowchart of a power plant anomaly detection method in an embodiment of the present application; Figure 7 is another flowchart of a power plant anomaly detection method in an embodiment of the present application; Figure 8 is a system block diagram of a power plant anomaly detection method in an embodiment of the present application; Figure 9 is a principle block diagram of the power station anomaly detection device in an embodiment of the present application; Figure 10 is a principle block diagram of the computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0017] In an embodiment, as shown in Figure 1 , a power station anomaly detection method is provided, comprising the following steps: S101, determining a reference power station and a to-be-detected power station other than the reference power station from a distributed photovoltaic power station system; S102, obtaining normalized power daily curves of the to-be-detected power station and the reference power station, and clustering the to-be-detected power station according to the normalized power daily curves to obtain a plurality of clustering clusters; S103, determining a first candidate low-power power station and a first confidence corresponding to the first candidate low-power power station from the to-be-detected power station based on a normalized power daily mean curve of each clustering cluster and the normalized power daily curve of the reference power station; S104, determining a power loss rate of the to-be-detected power station compared with the reference power station based on the normalized power daily curves of the to-be-detected power station and the reference power station, and determining a second candidate low-power power station and a second confidence corresponding to the second candidate low-power power station from the to-be-detected power station according to the power loss rate; S105, determining a target low-power power station based on the first candidate low-power power station and the first confidence, and the second candidate power station and the second confidence; S106, obtaining an inverter DC side input power, an AC three-phase voltage and an inverter safety voltage of the target low-power power station, and determining an anomaly evaluation result of the target low-power power station based on the inverter DC side input power, the AC three-phase voltage and the inverter safety voltage.

[0018] As an example, in step S101, a photovoltaic power station in a distributed photovoltaic power station system that needs to be detected for power station anomaly can be obtained as a to-be-detected power station, and a plurality of photovoltaic power station data of the same region, the same type and normal operation as the to-be-detected power station are obtained, sorted from high to low according to power generation, and the photovoltaic power station meeting a preset selection condition (for example, ranking at 40%, but not limited to this) is taken as a reference power station, so as to realize the acquisition of the reference power station and the to-be-detected power station.

[0019] As an example, in step S102, DC side power data (i.e. inverter DC side input power) of the to-be-detected power station and the reference power station can be obtained, and the DC side power data is normalized according to the installed capacity of the to-be-detected power station and the reference power station to obtain the normalized power daily curve of each power station under the same time scale.

[0020] Further, the normalized power daily curve can be subjected to cluster analysis, for example using K-means or hierarchical clustering algorithm, to cluster the to-be-detected power plants and divide them into groups with similar operation modes, thereby obtaining multiple clusters for subsequent differential analysis.

[0021] As an example, in step S103, the to-be-detected power plants in each cluster are subjected to mean processing to obtain the normalized power daily mean curve of the to-be-detected power plants in each cluster, and the mean curve is compared with the normalized power daily curve of the reference power plant, so that the to-be-detected power plants in the cluster corresponding to the normalized power daily mean curve lower than the reference are taken as the first candidate low-power power plants, and the first confidence is determined according to the normalized daily mean curve of the first candidate low-power power plants.

[0022] The normalized power daily mean curve is composed of normalized power daily means at multiple sampling times, and the normalized power daily mean at each sampling time is the arithmetic mean of the normalized power of all to-be-detected power plants in the cluster at that time, which is used to reflect the overall power level of the cluster at the corresponding period. Therefore, when determining the first candidate low-power power plants, the normalized power daily mean at each sampling time can be compared with the normalized power value of the reference power plant at the corresponding time point by point, the cluster with all normalized power daily means lower than the normalized power value is determined as a low-power cluster, and the to-be-detected power plants in the cluster are determined as the first candidate low-power power plants; or the cluster with the proportion of sampling points whose normalized power daily mean is lower than the normalized power value of the reference power plant at the corresponding time point exceeding a preset threshold (such as 50%) can be determined as a low-power cluster, and the to-be-detected power plants contained therein are marked as the first candidate low-power power plants, but not limited thereto, and the determination method is not limited by the present disclosure.

[0023] As an example, in step S104, the power difference value at each sampling time can be calculated based on the deviation between the normalized power daily mean curve of the first candidate low-power power plant and the reference power plant curve, thereby obtaining the power loss rate of the to-be-detected power plant compared with the reference power plant.

[0024] Further, the to-be-detected power plants with a power loss rate not lower than a preset power loss threshold can be screened out by comparing the power loss rate of the to-be-detected power plant with the preset power loss threshold, and determined as the second candidate low-power power plants, and the second confidence is determined according to the power loss rate of the second candidate low-power power plants.

[0025] As an example, in step S105, the first candidate low-power power station can be directly determined as the target low-power power station, the second candidate low-power power station can be directly determined as the target low-power power station, or the first candidate low-power power station and the second candidate low-power power station can be screened by using the first confidence and the second confidence to determine the target low-power power station.

[0026] Specifically, the first confidence and the second confidence can be combined for weighted fusion, and the fusion confidence of the first candidate low-power power station and the second candidate low-power power station is compared with the preset confidence threshold to determine the target low-power power station.

[0027] For example, by determining the intersection of the first candidate low-power power station and the second candidate low-power power station, the photovoltaic power station in the intersection whose fusion confidence is greater than the preset confidence threshold is determined as the target low-power power station.

[0028] It should be understood that the first candidate low-power power station and the second candidate low-power power station can not completely overlap, so in the fusion confidence calculation process, for the to-be-detected power station that exists in the first candidate low-power power station but is not determined as the second candidate low-power power station, the second confidence thereof can be set as a default minimum value or zero; for the to-be-detected power station that exists in the second candidate low-power power station but is not determined as the first candidate low-power power station, the first confidence thereof can be set as a default minimum value or zero, to ensure the integrity and fairness of the fusion confidence calculation.

[0029] As an example, in step S106, the related operation data of the target low-power power station can be obtained, including the inverter DC side input power, the AC three-phase voltage and the inverter safety voltage, so as to analyze the operation data to determine the abnormal reason and the abnormal time of the target low-power power station, and generate the corresponding abnormal evaluation result.

[0030] In summary, the power station abnormality detection method provided by the present disclosure is suitable for a distributed photovoltaic power station system, and includes: determining a reference power station and a to-be-detected power station other than the reference power station from the distributed photovoltaic power station system; obtaining a normalized power daily curve of the to-be-detected power station and the reference power station, and clustering the to-be-detected power station according to the normalized power daily curve to obtain a plurality of clustering clusters; determining a first candidate low-power power station from the to-be-detected power station based on a normalized power daily mean curve of each clustering cluster and the normalized power daily curve of the reference power station, and a first confidence degree corresponding to the first candidate low-power power station; determining a power loss rate of the to-be-detected power station compared with the reference power station based on the normalized power daily curves of the to-be-detected power station and the reference power station, and determining a second candidate low-power power station from the to-be-detected power station according to the power loss rate, and a second confidence degree corresponding to the second candidate low-power power station; determining a target low-power power station based on the first candidate low-power power station and the first confidence degree, and the second candidate low-power power station and the second confidence degree; obtaining an inverter DC side input power, an AC three-phase voltage, and an inverter safety voltage of the target low-power power station, and determining an abnormality evaluation result of the target low-power power station based on the inverter DC side input power, the AC three-phase voltage, and the inverter safety voltage. The method of the present disclosure effectively improves the accuracy and robustness of the identification of the low-power abnormal power station through a double candidate screening mechanism and a confidence degree weighted fusion strategy, avoids the misjudgment risk caused by a single index judgment, and simultaneously, through the deep analysis of the multi-dimensional operation data of the inverter, not only can the abnormality cause be accurately located, but also the timing characteristics of the abnormality can be accurately described, thereby providing a reliable basis for subsequent operation and maintenance decisions.

[0031] In an optional embodiment, when the target low-power power station and the abnormality evaluation result are determined, the acquired related operation data can be processed through the following preprocessing mode, so as to further improve the identification accuracy of the target low-power power station and the reliability of the abnormality evaluation result.

[0032] Specifically, when the related data (such as the normalized power daily curve, the inverter DC side input power, the AC three-phase voltage, and the inverter safety voltage, but not limited thereto) of the to-be-detected power station and the reference power station are acquired, the related data can be standardized, the original data collected is uniformly converted, including standardizing the time field, and multi-level sorting according to the power station identifier and the inverter identifier, reconstructing the time sequence structure, and ensuring the time consistency of the subsequent analysis.

[0033] In view of the missing values and abnormal values in the related data, the sliding window mean method, the median interpolation method, and the dynamic interpolation method based on the time sequence prediction model (such as LSTM and Prophet) are used to ensure the integrity and continuity of the input data; the obviously abnormal points are removed or replaced by the threshold clipping or 3σ method based on the multi-dimensional feature distribution.

[0034] For the case that the input power field (such as the inverter DC side input power) is missing or abnormal, the DC side power of the power station at time t can be estimated based on the voltage and current data of each maximum power point tracking (MPPT) channel, which is calculated by multiplying and summing each channel. The calculation formula is as follows: wherein, P k (t) represents the DC side power of the kth channel at time t (i.e. the inverter DC side input power described above), and K represents the number of MPPT channels; V k (t) represents the voltage of the kth channel at time t, I k (t) represents the current of the kth channel at time t.

[0035] For the case that the sampling time is irregular when the related data is sampled, resampling and time alignment can be performed according to the preset time granularity (5 min) to generate regularized timestamps, ensuring that each power station has comparability at the same time point. During the resampling process, interpolation or filling strategies can be used to fill in the missing data, and the sampling points exceeding the time granularity can be merged according to the nearest time granularity point to ensure the uniformity and continuity of the time series.

[0036] At the same time, meteorological data (including environmental temperature, solar radiation intensity, solar elevation angle, etc.) can be obtained in combination with the geographical location information (latitude and longitude or detailed address description) of the power station, and the missing values and outliers can be filled and processed after the timestamps are unified; and the meteorological sequence is resampled according to the preset time granularity (such as 5 min) and aligned with the power generation sequence (such as the inverter DC side input power described above) for subsequent analysis.

[0037] In an optional embodiment, after determining the reference power station and the power station to be detected, it can be determined whether to start the abnormal diagnosis and detection of the power station to be detected according to the full-load hours and equipment conversion rate of the reference power station, so as to avoid false judgment under low irradiation conditions.

[0038] Specifically, when the full-load hours of the reference power station are greater than the preset time limit (such as 1 hour) or the equipment conversion rate is higher than the preset conversion rate threshold (such as 70%), it is considered to have good irradiation conditions, and the system enters the abnormal reason diagnosis process; otherwise, it is determined that the weather condition is insufficient, and the abnormal diagnosis is suspended to avoid false judgment under low irradiation conditions.

[0039] Optionally, the calculation formula of the reference power station equipment conversion rate is as follows: wherein, η represents the equipment conversion rate (%, , denotes the reference power station daily full load hours, denotes the reference power station tilt surface irradiation hours.

[0040] In an embodiment, for the acquisition of the normalized power daily curve, the inverter DC side input power of the to-be-detected power station and the reference power station can be respectively unitized according to the photovoltaic installed capacity of the to-be-detected power station and the reference power station to obtain the dimensionless normalized power of the to-be-detected power station and the reference power station at time t, thereby forming the normalized power daily curve, and the calculation formula is as follows: wherein, denotes the normalized power (which can be the normalized power of the to-be-detected power station or the normalized power of the reference power station, depending on whether the parameter source is the reference power station or the to-be-detected power station), denotes the photovoltaic installed capacity (kW).

[0041] In an example, as shown in Figure 2 step S102, i.e., clustering the to-be-detected power station according to the normalized power daily curve, obtains a plurality of clustering clusters, including: S201, sampling the normalized power daily curve based on a first preset time granularity to determine the daily operation feature vector of the to-be-detected power station; S202, clustering the to-be-detected power station based on the daily operation feature vector to obtain a plurality of clustering clusters.

[0042] As an example, in step S201, the normalized power curve of the power station within a day can be sampled at a fixed time granularity (i.e., the first preset time granularity described above) to form a daily operation feature vector, and meteorological parameters (irradiance, temperature) can be additionally used as auxiliary features.

[0043] wherein the daily operation feature vector expression can be: , T denotes the number of sampling points in a day; denotes the normalized power of the i-th power station (including the reference power station and the to-be-detected power station) at time .

[0044] As an example, in step S202, a clustering algorithm such as K-means or hierarchical clustering can be used to cluster the to-be-detected power station, and power stations with similar daily operation features are divided into the same cluster, thereby identifying a power station group with similar operation characteristics.

[0045] Specifically, taking the K-means unsupervised clustering algorithm as an example, the normalized power daily curves of a plurality of to-be-detected power stations are divided into K clusters, and the optimal clustering center is obtained by minimizing the within-cluster sum of squares, and the expression of the optimal clustering center is: Where K represents the total number of clusters; Let represent the sample set of the k-th cluster; This represents the center curve of the k-th cluster.

[0046] In one instance, such as Figure 3 As shown, step S103, which involves determining a first candidate low-power power station from the power stations to be tested, based on the normalized daily power average curve of each cluster and the normalized daily power curve of the reference power station, and the first confidence level corresponding to the first candidate low-power power station, includes: S301, based on the normalized daily power curve corresponding to each cluster, perform mean processing to determine the normalized daily power mean curve for each cluster; S302, compare the normalized daily power average curve with the normalized daily power curve of the reference power plant to determine the relative power ratio of each cluster; S303, by comparing the preset relative power ratio threshold with the relative power, the power plants to be tested corresponding to the clusters whose relative power ratio is less than the preset relative power ratio threshold are identified as the first candidate low power power plants, and the relative power ratio corresponding to the first candidate low power power plants is identified as the first confidence level.

[0047] As an example, in step S301, the normalized daily power curves of all power plants to be tested within each cluster can be averaged point by point to obtain the normalized daily power average curve of that cluster. The calculation formula is as follows: in, This represents the normalized power of the i-th power station at time t. This represents the average normalized power of the k-th cluster at time t, and thus the daily average normalized power curve of the cluster is constructed from the average normalized power at each time.

[0048] As an example, in step S302, the daily average curve of the reference power plant can be used as a reference, and the relative power ratio of each cluster is defined, and the calculation formula is as follows: in, This represents the relative power ratio of the k-th cluster. This represents the normalized power of the reference power plant at time t. To prevent division by zero and avoid a denominator of zero, a sufficiently small value that will not significantly affect the result is generally chosen.

[0049] As an example, in step S303, the relative power ratio threshold can be preset, and the relative power ratio of the cluster corresponding to the power station to be detected is compared with the preset relative power ratio threshold. If the relative power ratio of the cluster is less than the preset relative power ratio threshold, the power station to be detected corresponding to the cluster is determined as the first candidate low-power power station, and the relative power ratio of the first candidate low-power power station is determined as the first confidence degree.

[0050] In other words, if the relative power ratio of a cluster is less than the preset threshold, it can be determined that all power stations in the cluster are the first candidate low-power power stations, and the relative power ratio is the corresponding first confidence degree. The lower the confidence degree, the more the power performance deviates from the normal level, and the higher the abnormality possibility.

[0051] In an example, as shown in FIG. 1 1, step S104 includes determining the power loss rate of the power station to be detected relative to the reference power station based on the normalized power daily curve of the power station to be detected and the reference power station, and determining the second candidate low-power power station from the power station to be detected and the second confidence degree corresponding to the second candidate low-power power station according to the power loss rate. Figure 4 S401, sampling the normalized power daily curve based on a second preset time granularity to obtain the resampled normalized power of the power station to be detected and the reference power station; S402, determining the power loss rate of the power station to be detected and the reference power station at a plurality of sampling time points based on the resampled normalized power; S403, comparing the preset power loss threshold with the power loss rate to determine the abnormal time points at which the power loss rate is greater than or equal to the preset power loss threshold; S404, determining the power station to be detected as the second candidate low-power power station if the number of abnormal time points is greater than or equal to a preset abnormal number threshold, and determining the ratio of the number of abnormal time points of the second candidate low-power power station to the total number of sampling time points as the second confidence degree.

[0052] As an example, in step S401, the normalized power daily curve can be sampled at a second preset time granularity (for example, 15 minutes) to obtain the resampled normalized power of the power station to be detected and the reference power station at the same time, so as to eliminate the error caused by the difference in sampling frequency. The second preset time granularity can be the same as or different from the first preset time granularity, which is not limited in the present disclosure.

[0053] As an example, in step S402, the power loss rate of the power station to be detected relative to the reference power station is determined by calculating the power normalized power ratio of the power station to be detected and the reference power station at each sampling time point, which reflects the power deviation degree of the power station to be detected at the corresponding time. For example, the calculation formula is: wherein, ​a power loss rate of the power station i to be detected at time t; a resampled normalized power of the power station to be detected at time t, a resampled normalized power of the reference power station at time t; a maximum value of a and b.

[0054] As an example, in step S403, the sampling time with the power loss rate greater than or equal to the preset power loss threshold can be determined as the abnormal time by comparing the preset power loss threshold with the power loss rate, so as to identify the specific time of the abnormal operation of the power station to be detected.

[0055] As an example, in step S404, the power station to be detected with the number of abnormal times greater than or equal to the preset abnormal number threshold can be determined as the second candidate low-power power station, and the ratio of the number of abnormal times of the second candidate low-power power station to the total number of sampling times can be determined as the second confidence.

[0056] In other words, by counting the abnormal times of the power station to be detected, the power abnormal frequency of the power station to be detected is determined, and then whether it is abnormal is evaluated.

[0057] As an example, the daily power generation period of the power station to be detected can be evenly divided into three periods, the proportion of continuous abnormalities (i.e. adjacent two sampling times are abnormal times) of each power station in each period p is calculated, and the calculation formula is as follows: wherein d is a date; a abnormality proportion of the power station to be detected at the date and the period; a number of abnormal frames in the period; a total sampling frame number in the period. For example, the abnormality proportion threshold is set as If the power station to be detected satisfies in any period, it is determined that the power station is the second candidate low-power power station at the date d, and the ratio between the total abnormal frame number and the total sampling frame number of the second candidate low-power power station in the three periods is determined as the second confidence.

[0058] It should be understood that the above is only an example, and the daily power generation period of the power station to be detected can also be evenly divided into more periods, or the period division can be dynamically adjusted according to the actual light intensity change, or the period division can not be performed, and the abnormality proportion is directly counted based on the whole period to determine the second candidate low-power power station.

[0059] In an example, as shown in Figure 5 step S105, i.e. determining the target low-power power station based on the first candidate low-power power station and the first confidence, and the second candidate power station and the second confidence, comprises: S501, based on the first confidence level and the first preset weight corresponding to the first confidence level, and the second confidence level and the second preset weight corresponding to the second confidence level, a weighted sum is performed to determine the fusion confidence level of the first candidate low-power power plant and the candidate second low-power power plant; S502, based on the fusion confidence level, compare it with the preset confidence threshold, and determine the first candidate low-power power plant and the second candidate low-power power plant whose fusion confidence level is greater than or equal to the preset confidence threshold as the target low-power power plant.

[0060] As an example, in step S501, the first confidence level and the second confidence level can be assigned a first preset weight and a second preset weight, respectively, to perform a weighted summation of the first confidence level and the second confidence level to obtain the fused confidence level, thereby achieving the fusion of fixed threshold comparison and cluster analysis. The fusion formula is as follows: in, Indicates the overall confidence level after fusion; Indicates the second confidence level; Indicates the first confidence level; Indicates the second preset weight, 1- This indicates the second preset weight. .

[0061] Furthermore, the power plant under test is judged based on the fusion confidence level and the preset confidence threshold, thereby identifying the target low-power power plant. The specific judgment formula is as follows: This indicates a preset reliability threshold, typically set to 0.5 as the equilibrium point, but is not limited to this. This indicates the result of the judgment. When the fusion confidence level is greater than or equal to the preset confidence threshold, the power station under test is determined to be the target low-power power station.

[0062] In one instance, such as Figure 6 As shown, step S106, which involves determining the anomaly assessment results of the target low-power power plant based on the inverter's DC-side input power, AC three-phase voltage, and inverter safety voltage, includes: S601, based on the DC-side input power of the inverter, determine the input power stop time of the target low-power power station; S602, based on the AC three-phase voltage and the inverter safety voltage, determines the critical overvoltage moment and critical peak-shaving moment of the target low-power power plant; S603, the moment when the input power stops and the critical overvoltage moment are the same is determined as the grid overvoltage moment, and the moment when the input power stops and the critical peak shaving moment are the same is determined as the grid peak shaving moment; S604, when the continuous time of the grid overvoltage moment is greater than or equal to the preset overvoltage duration threshold, the abnormal assessment result is determined to be that the target low-power power station has a grid overvoltage anomaly; S605, when the continuous time of the grid peak shaving moment is greater than or equal to the preset peak shaving duration threshold, the abnormal assessment result is determined to be that the target low-power power plant has a grid peak shaving abnormality.

[0063] As an example, in step S601, the acquired inverter DC-side input power can be statistically analyzed to determine the time corresponding to when the inverter DC-side input power is 0, and this time can be defined as the input power stop time. The expression for this is, for example,... .

[0064] As an example, in step S602, the moment when the target low-power power station is in a critical overvoltage state can be determined by comparing the AC three-phase voltage and the inverter safety voltage, and the moment when it is in a critical peak-shaving state can be determined by statistical analysis of the AC three-phase voltage.

[0065] Specifically, the critical overvoltage moment can be determined when the AC voltage of any phase of the three-phase AC voltage is greater than or equal to the difference between the inverter safety voltage and the preset tolerance. The specific formula is as follows: in, Indicates tolerance, used to account for measurement fluctuations. The x-phase AC voltage represents the three-phase AC voltage.

[0066] Specifically, the critical peak-shaving moment can be determined when the maximum voltage value of the three-phase AC voltage is approximately 0. A specific expression is as follows: .

[0067] As an example, in step S603, the same moment between the input power stop time and the critical overvoltage time can be determined as the grid overvoltage moment. That is, only when the output power is 0 and the critical overvoltage state is the moment determined as the grid overvoltage moment. This is to avoid errors in judging the grid overvoltage moment when the grid voltage is normal but the output power of the target low-power power station inverter is abnormal and the output power is 0. The specific formula is as follows: in, This indicates the moment when the power grid experiences overvoltage.

[0068] Meanwhile, the same time point in the input power stop time and the critical peak shaving time can be determined as the grid peak shaving time point, that is, when the output power is 0 and in the critical peak shaving state, the time point can be determined as the grid peak shaving time point, so as to avoid the error in the judgment of the grid peak shaving time point caused by the output power of 0 due to the abnormality of the inverter of the target low-power power station under the normal grid voltage, and the specific formula is, for example: wherein, represents the grid peak shaving time point.

[0069] As an example, in step S604, when the continuous time of the grid overvoltage time point is greater than or equal to the preset overvoltage duration threshold, it indicates that the target low-power power station has a persistent grid overvoltage state, and therefore, the target low-power power station has a grid overvoltage abnormality.

[0070] Further, in order to improve the identification accuracy of the grid overvoltage abnormality, the grid overvoltage time point, the grid overvoltage duration, and the specific phase set in which the overvoltage occurs can also be recorded, and a binary classification machine learning model (such as a LightGBM model, an XGBoost model, etc.) is introduced, taking the voltage distribution quantile of the grid overvoltage time point and the grid overvoltage duration as input features, to predict the grid overvoltage probability of the target low-power power station , and when , it is determined that the target low-power power station has a grid overvoltage abnormality, wherein is a preset overvoltage probability threshold.

[0071] As an example, in step S604, when the continuous time of the grid peak shaving time point is greater than or equal to the preset peak shaving duration threshold, it indicates that the target low-power power station has a persistent peak shaving behavior, and it is determined that the target low-power power station has a grid peak shaving abnormality.

[0072] Further, in order to improve the accuracy of the identification of the grid peak shaving abnormality, a binary classification machine learning model can be introduced, and the above-mentioned method for identifying the grid overvoltage abnormality can be used to review the judgment result to determine the target low-power power station, which will not be described here.

[0073] In an optional embodiment, when determining the abnormality evaluation result of the target low-power power station, a priority judgment principle can be used, that is, the identification of the "grid overvoltage" and "grid peak shaving" abnormalities is given priority; if any of the above types of abnormalities is detected on the day, it is determined that the operation state of the day is affected by a systematic external factor, and the detection of the subsequent abnormality type is not carried out, so as to avoid the misjudgment caused by external interference. If none of the above abnormalities is detected, the identification and analysis of the "string abnormality" and "flexible power limit" are continued, and the specific steps are as shown in Figure 7 , including: S701, obtaining the multi-path controller output current and the string specification of the target low-power power station when the target low-power power station does not have grid overvoltage abnormality and grid peak regulation abnormality; S702, determining the actual current ratio of each path of the controller based on the multi-path controller output current and the string specification of the power station group, and determining the time point when each actual current ratio is greater than or equal to a preset current ratio threshold as a string abnormality time point; S703, determining the input power peak value and the input power median value of the target low-power power station in a preset sampling interval based on the input power of the DC side of the inverter, to obtain the input power median absolute deviation of the target low-power power station; S704, when the product of the input power peak value and a preset deviation threshold is greater than the input power median absolute deviation, determining that the abnormality evaluation result is that the target low-power power station has flexible power limitation; S705, when the number of time points of the string abnormality time point is greater than or equal to a preset string abnormality threshold, determining that the abnormality evaluation result is that the target low-power power station has string abnormality.

[0074] As an example, in step S701, when the target low-power power station does not have grid overvoltage abnormality and grid peak regulation abnormality, it is indicated that the operation limitation of the target low-power power station is not dominated by external grid factors, and the running state of the equipment itself can be further investigated. At this time, the multi-path controller output current and the string specification are obtained for analyzing the abnormality reason of the target low-power power station.

[0075] As an example, in step S702, the actual current ratio of each path can be calculated based on the multi-path controller output current and the string specification. If the actual current ratio of a certain path is greater than or equal to a preset current ratio threshold, it indicates that the corresponding string may have faults such as short circuit or hot spot, and the time point is marked as a string abnormality time point.

[0076] Specifically, the actual current ratio can be determined by the following formula: wherein, is the output current of the kth MPPT path, is the actual current ratio vector at time t, which is used to represent the current running state of each MPPT branch.

[0077] Further, if the actual current ratio of any MPPT path deviates from the preset current ratio threshold, significantly, then the time point is determined as a string abnormality start time point, is the string specification stored in the configuration table of the target low-power power station.

[0078] As an example, in step S703, the target low-power power station can be determined based on the inverter DC side input power, the input power peak value and the input power median value of the preset sampling interval, and the input power median absolute deviation, i.e. the absolute value of the difference between the input power peak value and the input power median value, is calculated according to the following formula: wherein, represents a sliding window with i as the center and L as the length; represents the inverter DC side power of the jth sampling point; represents the inverter DC side power of the ith sampling point in the detection window; represents the input power median value of the power sequence in this period, represents the input power median absolute deviation of the ith sampling point.

[0079] wherein, when determining the input power median absolute deviation, the detection window can be divided into three sub-intervals according to hours, and the flexible limit power is independently detected in each sub-interval. In each sub-interval, the input power peak value is first calculated , and the data fluctuation caused by low light is screened out according to the peak value ratio, so as to avoid data interference, and the input power median absolute deviation of the center sliding window is calculated in the screened power sequence.

[0080] As an example, in step S704, when the product of the input power peak value and the preset deviation threshold is greater than the input power median absolute deviation, it is determined that the abnormal evaluation result is that the target low-power power station exists flexible limit power, i.e. the preset deviation threshold 1 , when , it is determined that the sampling point is in a stable platform segment. On this basis, the regression residual analysis method is used, and the solar irradiance, temperature and other factors are used as input to predict the theoretical inverter DC side input power, which is compared with the measured inverter DC side input power to assist in confirming the platform segment, thereby improving the identification accuracy of the low fluctuation limit power interval.

[0081] As an example, in step S705, when the number of group string abnormal time points is greater than or equal to the preset group string abnormal threshold, it is determined that the abnormal evaluation result is that the target low-power power station exists group string abnormality.

[0082] Optionally, the ratio between the number of group string abnormal time points and the total number of time points can also be used to determine whether the target low-power power station exists group string abnormality, for example, when the ratio between the number of group string abnormal time points and the total number of time points is 85%, it is marked as "group string all-day abnormality".

[0083] In an optional embodiment, when generating the abnormality evaluation result, detection and labeling of abnormal time can also be performed, such as continuous detection, start and end time and duration labeling, time period information generation, and additional information labeling.

[0084] Specifically, in the continuous detection, the sampling sequence of the single power station in time sequence is traversed, and the sampling points that continuously satisfy the same abnormal type judgment condition are merged into an abnormal candidate interval. Assuming that the sampling interval is , the minimum abnormal occurrence time is , and the number of continuous abnormal frames is , the following formula is satisfied: The candidate interval is determined as an effective abnormal period.

[0085] Specifically, in the start and end time and duration labeling, the effective abnormal period is recorded, the start time , the recovery time , and the duration . Among them, is the time of the last abnormal sampling point before the abnormality recovery.

[0086] Specifically, in the time period information generation, the is formatted as a time period representation of “HH:MM–HH:MM” as identification information of the abnormal period.

[0087] Specifically, in the additional information labeling, for the grid overvoltage period, the specific phase in which overvoltage occurs in the period is further counted and recorded.

[0088] For the grid peak shaving period, the string abnormality period, and the flexible power limiting period, the unified peak shaving / power limiting state is recorded, and no other information is involved.

[0089] The following is an exemplary description of the present disclosure, and the method of the present disclosure can include the following steps: 1. Data access and preprocessing: according to the photovoltaic power station operation data accessed by the system, unified formatting and data cleaning are performed, and corresponding power indicators are constructed to obtain a structured time series data set, ensuring data quality and consistency.

[0090] 2. Low-power station identification and screening: according to the preprocessed operation data, low-power identification indicators are constructed, combined with reference station data for comparison analysis and operation mode clustering analysis, and target power stations with persistent or phased low-efficiency performance are screened from the total power stations.

[0091] 3. Abnormal cause identification and classification diagnosis: For the screened low-power power station, based on the operating characteristic parameters and equipment state information of the power station, multi-source data fusion and machine learning model analysis are performed to identify the specific abnormal type of low-efficiency operation, and the abnormal type includes group string abnormality, flexible power limitation, power grid overvoltage, and power grid peak regulation.

[0092] 4. Abnormal period detection and labeling: In combination with the identified abnormal type and power fluctuation characteristics, time series analysis and change point detection are performed on the operating curve to automatically detect and label the corresponding abnormal duration, so as to realize the dual positioning of low-efficiency cause and actual impact period.

[0093] The specific implementation process of the method will be described in detail below with specific application scenarios as examples, which can include: S1: Access the operating data of the to-be-detected power station and its reference power station, including direct-current power, alternating-current voltage, current, daily power, and corresponding meteorological data. Standardize the original data format, unify the time field, and perform multi-level sorting, and process missing values and abnormal values; resample irregular sampling time points, and perform time alignment and granularity unification on the meteorological data; based on the processing result, construct normalized power, device conversion rate, and power loss rate characteristics.

[0094] S2: Determine whether to enter abnormal diagnosis according to the full-load hours and device conversion rate of the reference power station, and suspend if the conditions are insufficient to avoid low-irradiation interference; for the power station meeting the conditions, calculate the power loss rate and set a threshold to mark the preliminary abnormality, and determine the low-power day according to the abnormality proportion of daily power generation period. Extract the characteristics of the low-power day power curve and perform unsupervised clustering, and combine the threshold determination result to confirm the low-power abnormality.

[0095] S3: On the premise of confirming the low-power abnormality, identify and classify the causes of power grid overvoltage, power grid peak regulation, group string abnormality, and flexible power limitation in turn; for different abnormal types, combine rule determination and machine learning classification model to optimize the result, and improve the diagnosis accuracy and robustness.

[0096] S4: Time domain accurate positioning is performed on the identified abnormal event, and the abnormal start and end time, duration, and related characteristic information are recorded. The result is visualized and displayed, and is pushed to the operation and maintenance end to realize the operation and maintenance feedback closed-loop update of the diagnosis model.

[0097] The method is applied to Figure 8 The system is illustrated by taking the system shown in FIG. 1 as an example, and the system architecture includes but is not limited to a basic data layer, a business response layer, a core algorithm layer, and a system interaction layer.

[0098] S1: In the basic data layer, a power station information database, a weather information database, a diagnosis model database, a diagnosis result database, and a feedback information database are proposed.

[0099] S1.1: The power station information data stored in the power station information database specifically includes: power station number, inverter number, date, power station longitude and latitude coordinates, power station address, photovoltaic installed capacity, azimuth angle, inclination angle, inverter DC side power, MPPT current and voltage of each path, inverter three-phase voltage and current, daily power, full-load hour, reference power station number.

[0100] S1.2: The weather information data stored in the weather information database specifically includes: temperature, solar elevation angle, solar azimuth angle, global horizontal irradiance, direct solar irradiance, scattered horizontal irradiance, global tilt irradiance.

[0101] S1.3: The diagnosis model database stores the results of the low-efficiency power station identification model and the abnormal reason diagnosis model obtained by training, and the result parameters of the continuously updated optimization model, specifically including: low-efficiency power station identification model parameters, abnormal reason diagnosis model parameters.

[0102] S1.4: The diagnosis result data stored in the diagnosis result database specifically includes: abnormal equipment number, abnormal date, abnormal start and end time, abnormal duration, additional information label.

[0103] S1.5: The feedback information database is used to store the verification feedback information of the low-power power station abnormal diagnosis result of the operation and maintenance system.

[0104] S2: A preprocessing data management module, a diagnosis result management module and a feedback information management module are proposed in the business response layer.

[0105] S2.1: The preprocessing data management module is used to receive the cleaned power station operation data and meteorological data from the basic data layer, and provide input data for the low-efficiency power station identification module. This module supports batch operation of data and has data management capability to ensure the timeliness and consistency of business processing.

[0106] S2.2: The diagnosis result management module is used to support the system to call other modules to process the information of low-power power stations (such as target power stations) and generate abnormal diagnosis results, and present the results to the system, and store the diagnosis results in the basic data layer. In addition, the generated diagnosis results are uniformly managed, including query, screening and export of historical records, and labeling and maintaining the state (processed, to be verified, etc.) of the diagnosis results. This module supports integration with the permission management system to realize hierarchical access and operation control of the diagnosis results by different roles.

[0107] S2.3: A feedback information management module is used to support system operation and maintenance to verify and feedback the abnormal diagnosis results. The specific business process is that the system receives the feedback of the diagnosis results after the system operation and maintenance verifies the diagnosis results, and stores the feedback into the feedback information database. When the amount of system operation and maintenance feedback information reaches a certain standard, the system sends a feedback information reading instruction to the diagnosis model updating module.

[0108] S3: An original data processing module, an inefficient power station identification module, an abnormal reason diagnosis module and a diagnosis model updating module are proposed in the core function layer.

[0109] S3.1: The original data processing module is used to receive the power station operation data and meteorological data from the business response layer, and to perform feature extraction, time series alignment and normalization processing, thereby providing standardized input for subsequent low-power identification and abnormal diagnosis.

[0110] S3.2: The low-power station identification module is used to calculate power loss rate, device conversion rate and other indicators based on the standardized input data, and to screen out low-power running power stations and output identification results by combining the preset threshold and operation mode clustering analysis method.

[0111] S3.3: The abnormal reason diagnosis module is used to analyze the operation mode by calling the rule judgment and machine learning classification model in the low-power power station set, identify and classify the abnormal types, including grid overvoltage, grid peak shaving, string abnormality and flexible power limiting, and output the corresponding abnormal label and time period information.

[0112] S3.4: It is used to read the operation and maintenance feedback information and upload it to the server, and at the same time receive the optimized model parameters from the server, so as to realize the automatic updating and iteration of the local model, and improve the diagnosis accuracy and adaptability.

[0113] S4: A diagnosis result query interface and an operation and maintenance verification feedback interface are proposed in the system interaction layer.

[0114] S4.1: The diagnosis result query general interface is used to display the low-power power station list, including power station number, power station type, business order number, photovoltaic installed capacity, alarm state, power station address and detail operation. The interface supports query and screening by power station number, and provides a keyword search function. Users can click "Details" to view the abnormal detail page. The detail page of grid overvoltage and flexible power limiting is taken as an example. The detail page displays the original data curve, date, abnormal reason, abnormal start and end time, abnormal duration, daily power generation, daily full-load hour, etc. so as to enable the operation and maintenance personnel to quickly locate the problem.

[0115] S4.2: The operation and maintenance verification feedback interface is used for the operation and maintenance personnel to enter feedback information after verifying the abnormal diagnosis result on the spot. The feedback information is automatically associated with the corresponding diagnosis result record after being submitted and stored in the feedback information database. The system summarizes the feedback data accumulated to a certain amount in the background and automatically triggers the diagnosis model update module to read the feedback data for optimizing the model parameters and rule thresholds, thereby improving the accuracy and adaptability of subsequent diagnosis.

[0116] In summary, the present application proposes a low-power power station abnormal reason diagnosis method and system for distributed photovoltaic power stations. The method builds a unified diagnosis system suitable for multiple sites and multiple environments, integrates meteorological data, power station operation data and equipment state information, combines rule determination and intelligent algorithm model, realizes the automatic screening of low-power power stations, accurate classification of abnormal types and reason tracing analysis, and provides efficient support for operation and maintenance decision-making of distributed photovoltaic power stations.

[0117] The system adopts a multi-source data fusion and hierarchical analysis architecture, and organically connects data access and preprocessing, low-power identification, abnormal reason diagnosis, result verification and feedback modules. In the method, intelligent diagnosis means such as running mode clustering, power loss rate analysis and abnormal feature extraction are introduced, which not only retains the high interpretability of rule determination, but also improves the robustness and self-adaptive ability of abnormal identification in complex scenarios. The system can stably operate under different geographical locations and weather conditions, and realize the visualization of diagnosis results and operation and maintenance closed-loop feedback.

[0118] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0119] In an embodiment, an abnormal power station detection device is provided, which corresponds to the abnormal power station detection method in the above embodiment. As shown in the figure, the abnormal power station detection device includes a power station classification module 901, a data acquisition module 902, a first power station identification module 903, a second power station identification module 904, a target power station identification module 905, and an abnormality evaluation module 906. The functions of each module are described in detail as follows: Figure 9 The power station classification module 901 is used to determine a reference power station from a distributed photovoltaic power station system, and a power station to be detected other than the reference power station; The data acquisition module 902 is used to acquire the normalized power daily curve of the power station to be detected and the reference power station, and cluster the power station to be detected according to the normalized power daily curve to obtain a plurality of clustering clusters; ​The first power station identification module 903 is configured to determine a first candidate low-power power station from the to-be-detected power station based on the normalized power daily mean curve of each clustering cluster and the normalized power daily curve of the reference power station, and determine a first confidence degree corresponding to the first candidate low-power power station; The second power station identification module 904 is configured to determine a power loss rate of the to-be-detected power station compared with the reference power station based on the normalized power daily curves of the to-be-detected power station and the reference power station, and determine a second candidate low-power power station from the to-be-detected power station according to the power loss rate, and determine a second confidence degree corresponding to the second candidate low-power power station; The target power station identification module 905 is configured to determine a target low-power power station based on the first candidate low-power power station and the first confidence degree, and the second candidate low-power power station and the second confidence degree. The anomaly evaluation module 906 is configured to acquire an inverter direct-current side input power, an alternating-current three-phase voltage and an inverter safety voltage of the target low-power power station, and determine an anomaly evaluation result of the target low-power power station based on the inverter direct-current side input power, the alternating-current three-phase voltage and the inverter safety voltage.

[0120] In an embodiment, the data acquisition module 902 is further configured to sample the normalized power daily curve based on a first preset time granularity to determine a daily operation feature vector of the to-be-detected power station. The to-be-detected power station is clustered based on the daily operation feature vector to obtain a plurality of clustering clusters.

[0121] In an embodiment, the first power station identification module 903 is further configured to perform mean processing on the normalized power daily curve corresponding to each clustering cluster to determine a normalized power daily mean curve of each clustering cluster. The normalized power daily mean curve is compared with the normalized power daily curve of the reference power station to determine a relative power ratio of each clustering cluster. The to-be-detected power station corresponding to the clustering cluster with the relative power ratio less than the preset relative power ratio threshold is determined as the first candidate low-power power station by using the preset relative power ratio threshold and the relative power ratio comparison, and the relative power ratio corresponding to the first candidate low-power power station is determined as the first confidence degree.

[0122] In an embodiment, the second power station identification module 904 is further configured to sample the normalized power daily curve based on a second preset time granularity to acquire resampled normalized power of the to-be-detected power station and the reference power station. The power loss rates of the to-be-detected power station and the reference power station at a plurality of sampling time points are determined based on the resampled normalized power. The sampling time points with the power loss rate greater than or equal to a preset power loss threshold are determined as abnormal time points by using the preset power loss threshold and the power loss rate comparison. The power station to be detected with the number of abnormal moments greater than or equal to the preset abnormal number threshold is determined as a second candidate low-power power station, and a ratio of the number of abnormal moments of the second candidate low-power power station to the total number of sampling moments is determined as a second confidence degree.

[0123] In an embodiment, the target power station identification module 905 is further configured to determine a fusion confidence degree of the first candidate low-power power station and the second candidate low-power power station based on a weighted sum of the first confidence degree and a first preset weight corresponding to the first confidence degree, and the second confidence degree and a second preset weight corresponding to the second confidence degree. The first candidate low-power power station and the second candidate low-power power station with the fusion confidence degree greater than or equal to a preset confidence degree threshold are determined as the target low-power power station based on a comparison between the fusion confidence degree and the preset confidence degree threshold.

[0124] In an embodiment, the abnormality evaluation module 906 is further configured to determine an input power stop moment of the target low-power power station based on the input power of the DC side of the inverter. The critical overvoltage moment and the critical peak regulation moment of the target low-power power station are determined based on the three-phase AC voltage and the inverter safety voltage. The same moment in the input power stop moment and the critical overvoltage moment is determined as a grid overvoltage moment, and the same moment in the input power stop moment and the critical peak regulation moment is determined as a grid peak regulation moment. When the continuous time of the grid overvoltage moment is greater than or equal to a preset overvoltage duration threshold, it is determined that the abnormality evaluation result is that the target low-power power station has a grid overvoltage abnormality. When the continuous time of the grid peak regulation moment is greater than or equal to a preset peak regulation duration threshold, it is determined that the abnormality evaluation result is that the target low-power power station has a grid peak regulation abnormality.

[0125] In an embodiment, the abnormality evaluation module 906 is further configured to, when the target low-power power station does not have the grid overvoltage abnormality and the grid peak regulation abnormality, obtain a multi-channel controller output current and a power station group string specification of the target low-power power station. Based on the multi-channel controller output current and the power station group string specification, the actual current ratio of each channel of the controller is determined, and the moment when each actual current ratio is greater than or equal to a preset current ratio threshold is determined as a group string abnormal moment. Based on the input power of the DC side of the inverter, the input power peak value and the input power median value of the target low-power power station in a preset sampling interval are determined to obtain an input power median absolute deviation of the target low-power power station. When the product of the input power peak value and the preset deviation threshold is greater than the input power median absolute deviation, it is determined that the abnormality evaluation result is that the target low-power power station has flexible power limiting. When the number of abnormal string moments is greater than or equal to a preset string abnormal threshold, it is determined that the abnormal evaluation result is that the target low-power power station exists string abnormality.

[0126] The present application provides a kind of power station abnormality detection device, comprising: power station classification module, for determining reference power station from distributed photovoltaic power station system, and the power station to be detected except reference power station;Data acquisition module, for obtaining the normalized power daily curve of the power station to be detected and reference power station, and clustering the power station to be detected according to the normalized power daily curve, obtain a plurality of clustering clusters;First power station identification module, for determining the first candidate low-power power station based on the normalized power daily mean curve of each clustering cluster and the normalized power daily curve of reference power station, and the first confidence degree corresponding to the first candidate low-power power station;Second power station identification module, for determining the power loss rate of the power station to be detected compared with reference power station based on the normalized power daily curve of the power station to be detected and reference power station, and determining the second candidate low-power power station from the power station to be detected according to the power loss rate, and the second confidence degree corresponding to the second candidate low-power power station;Target power station identification module, for determining target low-power power station based on the first candidate low-power power station and the first confidence degree, and the second candidate power station and the second confidence degree;Abnormal evaluation module, for obtaining the input power of inverter DC side, AC three-phase voltage and inverter safety voltage of target low-power power station, and determining the abnormal evaluation result of target low-power power station based on the input power of inverter DC side, AC three-phase voltage and inverter safety voltage.The device of the present application, through double candidate screening mechanism and confidence weighted fusion strategy, effectively improves the accuracy and robustness of low-power abnormal power station identification, avoids the risk of misjudgment caused by single index determination, at the same time, combined with the depth analysis of inverter multidimensional operation data, not only can accurately locate abnormal reason, also can realize the accurate description of abnormal timing characteristics, provide reliable basis for subsequent operation and maintenance decision.

[0127] The device of the present application uniformly manages test software resources through a preset batch processing file, realizes the rapid loading and flexible calling of test program, and intuitively displays test items and instrument state through graphical interface, significantly improves the visualization degree and operation convenience of test process.

[0128] In one embodiment, a computer device, which can be a server, is provided, and an internal structure diagram thereof can be as shown in Figure 10The computer device includes a processor, a memory, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store data used in the power station anomaly detection method. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is configured to be executed by the processor to implement the power station anomaly detection method.

[0129] In one embodiment, a computer device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the power station anomaly detection method.

[0130] In one embodiment, a computer readable storage medium is provided, which stores a computer program executable by a processor to implement the power station anomaly detection method.

[0131] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium and can include the processes of the above-mentioned embodiments when executed. Any reference to memory, storage, database or other medium in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0132] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0133] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A power station anomaly detection method, applicable to a distributed photovoltaic power station system, characterized in that, The method comprises the following steps: determining a reference power station and a to-be-detected power station from the distributed photovoltaic power station system; obtaining normalized power daily curves of the to-be-detected power station and the reference power station, and clustering the to-be-detected power station according to the normalized power daily curves to obtain a plurality of clustering clusters; determining a first candidate low-power power station from the to-be-detected power station based on a normalized power daily mean curve of each clustering cluster and the normalized power daily curve of the reference power station, and a first confidence degree corresponding to the first candidate low-power power station; determining a power loss rate of the to-be-detected power station compared with the reference power station based on the normalized power daily curves of the to-be-detected power station and the reference power station, and determining a second candidate low-power power station from the to-be-detected power station according to the power loss rate, and a second confidence degree corresponding to the second candidate low-power power station; determining a target low-power power station based on the first candidate low-power power station and the first confidence degree, and the second candidate low-power power station and the second confidence degree; obtaining an inverter direct current side input power, an alternating current three-phase voltage and an inverter safety voltage of the target low-power power station, and determining an abnormality evaluation result of the target low-power power station based on the inverter direct current side input power, the alternating current three-phase voltage and the inverter safety voltage.

2. The method of claim 1, wherein, The clustering of the to-be-detected power station according to the normalized power daily curves to obtain a plurality of clustering clusters comprises the following steps: sampling the normalized power daily curves based on a first preset time granularity to determine a daily operation feature vector of the to-be-detected power station; clustering the to-be-detected power station based on the daily operation feature vector to obtain the plurality of clustering clusters.

3. The method of claim 1, wherein, The determination of a first candidate low-power power station from the to-be-detected power station based on a normalized power daily mean curve of each clustering cluster and the normalized power daily curve of the reference power station, and a first confidence degree corresponding to the first candidate low-power power station comprises the following steps: performing mean processing on the normalized power daily curve corresponding to each clustering cluster to determine a normalized power daily mean curve of each clustering cluster; comparing the normalized power daily mean curve with the normalized power daily curve of the reference power station to determine a relative power ratio of each clustering cluster; determining the to-be-detected power station corresponding to the clustering cluster with a relative power ratio less than a preset relative power ratio threshold as the first candidate low-power power station, and determining the relative power ratio corresponding to the first candidate low-power power station as the first confidence degree.

4. The method of claim 1, wherein, The determination of a power loss rate of the to-be-detected power station compared with the reference power station based on the normalized power daily curves of the to-be-detected power station and the reference power station, and the determination of a second candidate low-power power station from the to-be-detected power station according to the power loss rate, and a second confidence degree corresponding to the second candidate low-power power station comprises the following steps: sampling the normalized power daily curves based on a second preset time granularity to obtain resampled normalized powers of the to-be-detected power station and the reference power station; determine power loss rates of the to-be-detected power station and the reference power station at a plurality of sampling time points based on the resampled normalized power; determine, by comparing the power loss rates with a preset power loss threshold, sampling time points at which the power loss rates are greater than or equal to the preset power loss threshold as abnormal time points; determine, as the second candidate low-power power station, a to-be-detected power station at which the number of abnormal time points is greater than or equal to a preset abnormal time threshold, and determine, as the second confidence, a ratio of the number of abnormal time points of the second candidate low-power power station to the total number of sampling time points.

5. The method of claim 1, wherein, The determining, based on the first candidate low-power power station and the first confidence and the second candidate low-power power station and the second confidence, of the target low-power power station includes: performing weighted summation based on the first confidence, a first preset weight corresponding to the first confidence, the second confidence, and a second preset weight corresponding to the second confidence to determine a fusion confidence of the first candidate low-power power station and the second candidate low-power power station; and determine, as the target low-power power station, the first candidate low-power power station and the second candidate low-power power station whose fusion confidence is greater than or equal to a preset confidence threshold.

6. The method of claim 1, wherein, The determining, based on the inverter DC side input power, the AC three-phase voltage, and the inverter safety voltage, of the abnormal evaluation result of the target low-power power station includes: determining, based on the inverter DC side input power, an input power stop time point of the target low-power power station; determining, based on the AC three-phase voltage and the inverter safety voltage, a critical overvoltage time point and a critical peak regulation time point of the target low-power power station; determining, as a grid overvoltage time point, a same time point of the input power stop time point and the critical overvoltage time point, and determining, as a grid peak regulation time point, a same time point of the input power stop time point and the critical peak regulation time point; determining, as the abnormal evaluation result, that the target low-power power station has a grid overvoltage anomaly when a continuous time of the grid overvoltage time point is greater than or equal to a preset overvoltage duration threshold; and determining, as the abnormal evaluation result, that the target low-power power station has a grid peak regulation anomaly when a continuous time of the grid peak regulation time point is greater than or equal to a preset peak regulation duration threshold.

7. The method of claim 6, wherein, The method further includes: when the target low-power power station does not have the grid overvoltage anomaly and the grid peak regulation anomaly, obtaining a plurality of controller output currents of the target low-power power station and a power station group string specification; determining, based on the plurality of controller output currents and the power station group string specification, an actual current ratio of each path of a controller, and determining, as a group string abnormal time point, a time point at which the actual current ratio of each path of the controller is greater than or equal to a preset current ratio threshold; and determining, based on the inverter DC side input power, an input power peak value and an input power median value of the target low-power power station in a preset sampling interval to obtain an input power median absolute deviation of the target low-power power station. determining that the abnormality evaluation result is that the target low-power power station has flexible power limiting when the product of the input power peak value and the preset deviation threshold is greater than the input power median absolute deviation; determining that the abnormality evaluation result is that the target low-power power station has string abnormality when the number of times of the group string abnormality time is greater than or equal to the preset group string abnormality threshold.

8. A power plant abnormality detection device characterized by comprising: The method comprises: a power station classification module configured to determine a reference power station from the distributed photovoltaic power station system and a to-be-detected power station other than the reference power station; a data acquisition module configured to acquire normalized power daily curves of the to-be-detected power station and the reference power station, and cluster the to-be-detected power station according to the normalized power daily curves to obtain a plurality of clustering clusters; a first power station identification module configured to determine a first candidate low-power power station from the to-be-detected power station based on a normalized power daily mean curve of each of the clustering clusters and the normalized power daily curve of the reference power station, and a first confidence degree corresponding to the first candidate low-power power station; a second power station identification module configured to determine a power loss rate of the to-be-detected power station compared with the reference power station based on the normalized power daily curves of the to-be-detected power station and the reference power station, and determine a second candidate low-power power station from the to-be-detected power station according to the power loss rate, and a second confidence degree corresponding to the second candidate low-power power station; a target low-power power station identification module configured to determine a target low-power power station based on the first candidate low-power power station and the first confidence degree, and the second candidate low-power power station and the second confidence degree; an abnormality evaluation module configured to acquire an input power of a direct current side of an inverter, an alternating current three-phase voltage, and an inverter safety voltage of the target low-power power station, and determine an abnormality evaluation result of the target low-power power station based on the input power of the direct current side of the inverter, the alternating current three-phase voltage, and the inverter safety voltage.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the power station abnormality detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the power station abnormality detection method according to any one of claims 1 to 7.

Images