Photovoltaic power station three-dimensional digital twin modeling method and system based on point cloud scanning
Through high-precision point cloud scanning and dynamic data fusion, a digital twin of all elements of the photovoltaic power station is constructed, which solves the problems of insufficient accuracy and low efficiency in traditional photovoltaic power station modeling methods, realizes high-precision visualization and intelligent analysis of equipment status, and supports efficient and intelligent operation and maintenance of the power station.
Patent Information
- Application Number
- CN202510827806.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-19
- Publication Date
- 2025-10-17
AI Technical Summary
Traditional photovoltaic power station modeling methods have problems such as insufficient modeling accuracy and efficiency, static models, and high complexity in point cloud data processing, making it difficult to meet the needs of high-precision, high-efficiency and dynamic data fusion.
Through high-precision point cloud scanning, automated reverse modeling and dynamic data fusion, a full-factor digital twin of the photovoltaic power station is constructed to achieve equipment status visualization and intelligent analysis. The high-precision digital twin model is constructed by matching point cloud data density eigenvalues, calculating dynamic fluctuation indicators and generating model correction coefficients.
It achieves high-precision visualization and intelligent analysis of the equipment status of photovoltaic power stations, provides efficient digital support for the intelligent operation and maintenance of power stations, improves modeling accuracy and efficiency, and supports dynamic data fusion and fault diagnosis.
Smart Images
Figure CN120807776A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of photovoltaic power stations, in particular to a photovoltaic power station three-dimensional digital twin modeling method and system based on point cloud scanning. BACKGROUND
[0002] With the continuous expansion of photovoltaic power station scale and the increasing demand for intelligentization, high-precision and high-efficiency power station three-dimensional modeling technology has become the focus of the industry.
[0003] However, the traditional photovoltaic power station modeling method has the following technical bottlenecks: 1. Insufficient modeling accuracy and efficiency: Traditional methods mainly rely on manual mapping or low-resolution aerial imagery, resulting in low model accuracy and difficulty in capturing device details. In addition, manual mapping is inefficient and cannot meet the needs of rapid modeling of large-scale power stations.
[0004] 2. Static model, lack of dynamic data fusion: Existing digital twin models are mostly static three-dimensional displays and cannot be effectively integrated with real-time operation data of the power station, resulting in the model being unable to dynamically reflect the device operating state, limiting its application value in fault diagnosis, performance optimization, etc.
[0005] 3. High complexity of point cloud data processing: Although laser point cloud scanning technology can obtain high-precision power station three-dimensional data, existing algorithms are computationally complex in point cloud denoising, registration, segmentation, and semantic recognition, resulting in long modeling cycles and difficulty in adapting to the modeling needs of large-scale and high-frequency updates of photovoltaic power stations. SUMMARY
[0006] The present application provides a photovoltaic power station three-dimensional digital twin modeling method and system based on point cloud scanning, which constructs a full-factor digital twin body of the photovoltaic power station through high-precision point cloud scanning, automated reverse modeling, and dynamic data fusion, realizes device state visualization and intelligent analysis, and provides high-precision and efficient digital support for the intelligent operation and maintenance of the power station.
[0007] To achieve the above purpose, the present application provides a photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning, comprising: During the operation of the photovoltaic power station, the point cloud data and corresponding performance data at each scanning time within a predetermined time period before the current time are obtained by a three-dimensional scanning device, and the point cloud data density characteristic value of the scanning scene is collected; When the point cloud data density characteristic value is lower than the first density threshold, a historical point cloud missing distribution map and a current point cloud missing distribution map are generated, and a matching degree comparison is made with a predetermined reference missing distribution map, and a data collection anomaly alarm is triggered according to the matching result; According to the local change of each point cloud feature parameter and performance data at each scanning moment, combined with the dynamic compensation result of the point cloud data density characteristic value to the scanning quality, the dynamic fluctuation index of each point cloud feature parameter at the scanning moment is calculated; The influence weight of the point cloud feature parameter on the performance data is determined, the model correction coefficient at the scanning moment is generated based on the influence weight of each scanning moment and the current moment and the density compensation coefficient, and the digital twin model of the photovoltaic power station at the current moment is constructed based on the performance data and the model correction coefficient of all scanning moments in the preset time period.
[0008] Further, in the calculation of the dynamic fluctuation index of each point cloud feature parameter at the scanning moment according to the local change of each point cloud feature parameter and performance data at each scanning moment, combined with the dynamic compensation result of the point cloud data density characteristic value to the scanning quality, it includes: For any scanning moment, the geometric feature and reflectivity feature in the point cloud data of the arbitrary scanning moment are extracted as the feature parameters; When the light intensity and component temperature of the scanning scene exceed the threshold value, the compensation scanning configuration is started and the scanning stability period of the transition scanning area is calculated; A time window is set with the arbitrary scanning moment as the center, the standard deviation, the coefficient of variation and the local entropy value of each feature parameter and performance data in the time window are calculated combined with the density compensation coefficient, and the dynamic fluctuation index is generated.
[0009] Further, in the determination of the influence weight of the point cloud feature parameter on the performance data, it includes: In the preset time period, the differences of the point cloud feature parameters and the performance data in the dynamic fluctuation index and the density compensation coefficient are fused to determine the initial influence weight of each point cloud feature parameter on the performance data; According to the dynamic fluctuation correlation between each point cloud feature parameter and its initial influence weight, the final influence weight of each parameter on the performance data is calculated.
[0010] Further, in the fusion of the differences of the point cloud feature parameters and the performance data in the dynamic fluctuation index and the density compensation coefficient, the initial influence weight of each point cloud feature parameter on the performance data is determined, which includes: The Pearson correlation coefficient and time-lag mutual information of each point cloud feature parameter and performance data in the preset time period are calculated; The fluctuation time sequence deviation of the feature parameters and the performance data is quantified by the dynamic time warping algorithm; The Pearson correlation coefficient, time-lag mutual information and fluctuation time sequence deviation are weighted and fused with the density compensation coefficient, and the initial influence weight is obtained after normalization processing.
[0011] Further, in calculating the final influence weight of each parameter on performance data according to the dynamic fluctuation correlation between each point cloud feature parameter and its initial influence weight, it includes: Selecting the first K extreme points of each feature parameter dynamic fluctuation index as the key fluctuation point; At the key fluctuation point, combined with the data quality influence factor of the auxiliary scanning area, the fluctuation phase difference and amplitude ratio of other parameters and the target parameter are calculated; Based on the cross-frequency correlation analysis of Granger causality test and wavelet decomposition, the correlation correction factor is generated; The product of the correlation correction factor and the initial influence weight is taken as the final influence weight.
[0012] Further, in generating the model correction coefficient at the scanning time based on the point cloud feature parameter deviation degree at each scanning time and the current time, the final influence weight and the density compensation coefficient, it includes: Based on the Euclidean distance change rate and Hausdorff distance of the point cloud feature parameter, the space-time offset is calculated; Combined with the final influence weight and the density compensation coefficient, a Gaussian mixture model is constructed to output the correction coefficient at the scanning time; The Kalman filter is used for time sequence smoothing processing of the correction coefficient.
[0013] Further, in constructing the current time photovoltaic power station digital twin model based on the performance data and model correction coefficient of all scanning times within a predetermined time period, it includes: Based on the performance data, an initial photovoltaic power station digital twin model is constructed; Collecting the initial model parameters of the initial photovoltaic power station digital twin model, and correcting the initial model parameters based on the model correction coefficient to obtain the corrected initial model parameters; According to the corrected initial model parameters, a photovoltaic power station digital twin model is obtained.
[0014] Further, in collecting the initial model parameters of the initial photovoltaic power station digital twin model, and correcting the initial model parameters based on the model correction coefficient to obtain the corrected initial model parameters, it includes: Pre-set first and second preset model correction coefficients; Pre-set first, second and third preset adjustment coefficients; When the model correction coefficient is less than the first preset model correction coefficient, the product value of the first preset adjustment coefficient and the initial model parameter is calculated as the corrected initial model parameter; When the model correction coefficient is greater than or equal to the first preset model correction coefficient and less than the second preset model correction coefficient, a product value of the second preset adjustment coefficient and the initial model parameter is calculated as the corrected initial model parameter. When the model correction coefficient is greater than or equal to the second preset model correction coefficient, a product value of the third preset adjustment coefficient and the initial model parameter is calculated as the corrected initial model parameter.
[0015] Further, after constructing the digital twin model of the photovoltaic power station at the current time based on the performance data and the model correction coefficient at all scanning time points in the preset time period, the method further comprises: When the point cloud data is newly added by more than a threshold value or the density matching degree is abnormal, triggering the incremental training of the model; Based on the residual error between the output of the digital twin model of the photovoltaic power station and the actual data, a fault warning is performed, and a scanning quality abnormality warning is associated.
[0016] In order to achieve the above-mentioned purpose, the application further provides a photovoltaic power station three-dimensional digital twin modeling system based on point cloud scanning, comprising: A point cloud scanning module is configured to acquire point cloud data and corresponding performance data at each scanning time point in a preset time period before the current time through a three-dimensional scanning device during operation of the photovoltaic power station, and simultaneously acquire a point cloud data density characteristic value of a scanning scene; An abnormal matching module is configured to generate a historical point cloud missing distribution map and a current point cloud missing distribution map when the point cloud data density characteristic value is lower than a first density threshold value, and perform matching degree comparison with a preset reference missing distribution map, and trigger a data acquisition abnormality warning according to a matching result; An index calculation module is configured to calculate a dynamic fluctuation index of each point cloud characteristic parameter at a scanning time point according to a local change of each point cloud characteristic parameter and performance data at the scanning time point, and a dynamic compensation result of the scanning quality of the point cloud data density characteristic value; A model construction module is configured to determine an influence weight of the point cloud characteristic parameter on the performance data, generate a model correction coefficient at the scanning time point based on the influence weight at each scanning time point and the current time and a density compensation coefficient, and construct a digital twin model of the photovoltaic power station at the current time based on the performance data and the model correction coefficient at all scanning time points in the preset time period.
[0017] Compared with the prior art, the application has the following beneficial effects: The application acquires point cloud data and performance data, collects point cloud data density characteristic values; when the point cloud data density characteristic values are lower than a first density threshold, generates a historical point cloud missing distribution atlas and a current point cloud missing distribution atlas, triggers a data collection anomaly alarm according to a matching result; calculates a dynamic fluctuation index according to local changes of point cloud characteristic parameters and performance data; determines an influence weight of the point cloud characteristic parameters on the performance data, generates a model correction coefficient at a scanning moment, constructs a digital twin model of a photovoltaic power station based on the performance data and the model correction coefficient, constructs a full-factor digital twin body of the photovoltaic power station through high-precision point cloud scanning, automatic reverse modeling and dynamic data fusion, realizes equipment state visualization and intelligent analysis, and provides high-precision and high-efficiency digital support for intelligent operation and maintenance of the power station. BRIEF DESCRIPTION OF DRAWINGS
[0018] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not intended to limit the scope of the application. Moreover, the same reference numerals are used throughout the same figures. In the drawings: Figure 1 A flowchart of a photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning in an embodiment of the application is shown; Figure 2 A structural diagram of a photovoltaic power station three-dimensional digital twin modeling system based on point cloud scanning in an embodiment of the application is shown. DETAILED DESCRIPTION
[0019] The specific embodiments of the application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the application, but are not used to limit the scope of the application.
[0020] In the description of the present application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements indicated must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.
[0021] The terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise stated, the meaning of "multiple" is two or more.
[0022] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connecting" should be understood in a broad sense, for example, can be fixed connection, can also be detachable connection, or integral connection; can be mechanical connection, can also be electrical connection; can be direct connection, can also be indirect connection through an intermediate medium, can be internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.
[0023] The following is a description of the preferred embodiments of the present application in conjunction with the accompanying drawings.
[0024] As Figure 1 shown, the embodiments of the present application disclose a photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning, comprising: S110: In the operation process of the photovoltaic power station, the point cloud data and the corresponding performance data of each scanning time within a preset time period before the current time are acquired by a three-dimensional scanning device, and the point cloud data density characteristic value of the scanning scene is collected at the same time; In this embodiment, the three-dimensional scanning device includes a ground station scanner, an unmanned aerial vehicle scanner, etc., which is selected according to the actual situation.
[0025] In this embodiment, the preset time period is pre-set, such as 1 hour or 2 hours, etc.
[0026] In this embodiment, the point cloud data includes the three-dimensional coordinates, the reflection intensity and other characteristics such as the RGB color information of the surface of the photovoltaic module.
[0027] In this embodiment, the performance data includes output voltage, current, power, efficiency, etc.
[0028] In this embodiment, the point cloud density characteristic value of each scanning scene, such as a single photovoltaic module string or a specific area, is calculated to represent the data integrity.
[0029] S120: When the point cloud data density characteristic value is lower than the first density threshold value, generate a historical point cloud missing distribution map and a current point cloud missing distribution map, and compare the matching degree with a pre-set reference missing distribution map, and trigger a data collection abnormal alarm according to the matching result; In this embodiment, the first density threshold value is, for example, 500 points / m³.
[0030] In this embodiment, the historical point cloud missing distribution atlas is generated: the point cloud data of the scanning scene in the past N scanning periods (such as the last 24 hours) is called from the database. For each historical scanning time, the point cloud sparse area is detected, and its spatial position and missing degree are marked. The high-frequency occurrence position of the historical missing area (such as the component edge and the support shadow area) is counted. The historical missing data is projected into a two-dimensional heat map according to the arrangement direction of the photovoltaic array (such as the string direction), the horizontal axis is the position, the vertical axis is the time, and the color depth represents the missing frequency.
[0031] In this embodiment, the current point cloud missing distribution atlas is generated: for the current scanning frame, a grid (such as 0.5m*0.5m) is divided, and the DFV of each grid is calculated. All grids with DFV below the threshold are marked, and their spatial coordinates and density deviation values are recorded. The current missing area is aligned with the historical atlas to generate a current missing distribution atlas with the same coordinate system.
[0032] In this embodiment, the reference missing distribution atlas (preset reference) is generated: an ideal missing distribution is established in advance based on the physical characteristics of the equipment (such as the ranging attenuation curve of LiDAR) and the environmental model (such as atmospheric scattering). The farther away from the scanner, the lower the point cloud density, and the reference atlas reflects this attenuation rule. It includes the theoretical missing area (such as the component backboard and the support shielding area) and its allowed missing threshold.
[0033] In this embodiment, the matching degree comparison and abnormality determination are performed: the overlapping area ratio of the current / historical missing area and the reference atlas is calculated. If the ratio exceeds the preset ratio, an abnormal data acquisition alarm is triggered.
[0034] The beneficial effects of the above technical solutions are: the historical point cloud missing distribution atlas and the current point cloud missing distribution atlas are generated, and the matching degree comparison is performed with the preset reference missing distribution atlas. According to the matching result, an abnormal data acquisition alarm is triggered, which ensures the data acquisition accuracy and provides reliable data support for the three-dimensional digital twin modeling of the photovoltaic power station.
[0035] S130: According to the local change of each point cloud characteristic parameter and performance data at each scanning time, and the dynamic compensation result of the point cloud data density characteristic value to the scanning quality, the dynamic fluctuation index of each point cloud characteristic parameter at the scanning time is calculated; In some embodiments of the present application, when the dynamic fluctuation index of each point cloud characteristic parameter at the scanning time is calculated according to the local change of each point cloud characteristic parameter and performance data at each scanning time, and the dynamic compensation result of the point cloud data density characteristic value to the scanning quality, it includes: For any scanning time, the geometric feature and reflectivity feature in the point cloud data of the scanning time are extracted as the characteristic parameters; When the light intensity and component temperature of the scanned scene exceed the threshold value, a compensation scanning configuration is started and a scanning stability period of the transition scanning area is calculated; A time window is set with any scanning time as the center, standard deviations, coefficients of variation and local entropy values of each feature parameter and performance data in the time window are calculated in combination with the density compensation coefficient, and a dynamic fluctuation index is generated.
[0036] In this embodiment, the geometric features are: surface curvature: the local curvature of the point cloud is calculated by PCA to identify component deformation (such as bending caused by thermal expansion). Flatness deviation: a plane model is fitted, and the root mean square error (RMSE) of the point cloud to the plane is calculated to detect installation tilt or structural damage.
[0037] Reflectivity features: intensity mean / variance: LiDAR reflectivity intensity reflects surface material changes (such as dust accumulation leading to reduced reflectivity). Spectral analysis (if multi-band data is included): distinguish between dirt (low reflectivity) and shadows (high absorption).
[0038] In this embodiment, the component temperature exceeding the threshold value can be specifically set according to the operation of the component, which is not specifically limited here.
[0039] In this embodiment, the compensation scanning configuration: increase the number of scans: take the average of multiple scans of the same area. Adjust the laser power: increase the LiDAR emission intensity to penetrate the thermal halo effect in high-temperature environments.
[0040] In this embodiment, the scanning stability period: in the transition scanning area (such as the light / temperature gradient change area), the consistency of the point cloud of continuous scanning is calculated (measured by ICP registration error), and the stability period (such as Δt=5 minutes) is output.
[0041] In this embodiment, the time window: with the current scanning time t0 as the center, the window length T=10 minutes (including 5 minutes of data before and after) is selected.
[0042] The beneficial effects of the above technical solutions are: the present application sets a time window with any scanning time as the center, calculates the standard deviation, coefficient of variation and local entropy value of each feature parameter and performance data in the time window in combination with the density compensation coefficient, generates a dynamic fluctuation index, realizes comprehensive analysis of point cloud data and performance data, and ensures the accuracy of subsequent modeling.
[0043] S140: Determine the influence weight of the point cloud feature parameter on the performance data, generate a model correction coefficient at the scanning time based on the influence weight of each scanning time and the current time and the density compensation coefficient, and construct a digital twin model of the photovoltaic power station at the current time based on the performance data and the model correction coefficient of all scanning times in the preset time period.
[0044] In some embodiments of the present application, in determining the influence weight of the point cloud feature parameter on the performance data, the following steps are included: In a preset time period, the differences between the point cloud feature parameters and the performance data in the dynamic fluctuation index and the density compensation coefficient are fused to determine the initial influence weight of each point cloud feature parameter on the performance data. According to the dynamic fluctuation correlation between the point cloud feature parameters and the initial influence weight, the final influence weight of each parameter on the performance data is calculated.
[0045] In some embodiments of the present application, in fusing the differences between the point cloud feature parameters and the performance data in the dynamic fluctuation index and the density compensation coefficient to determine the initial influence weight of each point cloud feature parameter on the performance data, the following steps are included: Calculate the Pearson correlation coefficient and time-lag mutual information of each point cloud feature parameter and the performance data in a preset time period; Quantify the fluctuation time sequence deviation of the feature parameters and the performance data by dynamic time warping algorithm; Weighted fusion of the Pearson correlation coefficient, time-lag mutual information and fluctuation time sequence deviation and density compensation coefficient, and normalization processing to obtain the initial influence weight.
[0046] In this embodiment, the linear correlation between the point cloud features (such as curvature, reflectivity) and the performance data (such as power, efficiency) is calculated, and the value range is [-1, 1]. Positive value indicates that the feature and the performance change in the same direction, and negative value indicates that the feature and the performance change in the opposite direction.
[0047] In this embodiment, the time-lag mutual information: analyzes the nonlinear correlation of the feature and the performance data under different time delays, and captures the lagging influence (such as dust accumulation leading to delayed power decline).
[0048] In this embodiment, dynamic time warping (DTW): aligns the fluctuation curves of the feature and the performance data, calculates the time sequence offset distance, and quantifies the difference in change rhythm of the two (such as point cloud deformation earlier than power anomaly).
[0049] In this embodiment, weighted fusion and normalization: weighted fusion (weight adjustable) of the above three indexes and the density compensation coefficient (DCC) to enhance the contribution of high-quality data. Normalized to the range of [0, 1] to obtain the initial influence weight, which reflects the preliminary influence degree of each feature on the performance.
[0050] The beneficial effects of the above technical solutions are: the present application fuses the Pearson correlation coefficient, the time-lag mutual information and the fluctuation time sequence deviation with the density compensation coefficient, and obtains the initial influence weight after normalization processing, which provides a directional technical support for three-dimensional digital twin modeling of photovoltaic power station.
[0051] In some embodiments of the present application, in calculating the final influence weight of each parameter on performance data according to the dynamic fluctuation correlation between each point cloud feature parameter and its initial influence weight, it includes: Select the top K extreme points of each feature parameter dynamic fluctuation index as the key fluctuation point; At the key fluctuation point, combined with the data quality influence factor of the auxiliary scanning area, calculate the fluctuation phase difference and amplitude ratio of other parameters and the target parameter; Based on the cross-frequency correlation analysis of Granger causality test and wavelet decomposition, generate the correlation correction factor; The product of the correlation correction factor and the initial influence weight is taken as the final influence weight.
[0052] In this embodiment, the key fluctuation point is selected: the top K extreme points (such as peak / valley) of each feature dynamic fluctuation index (DFI) are extracted as the key fluctuation time, representing significant change events (such as sudden hot spots or shadow shielding).
[0053] In this embodiment, the phase difference and amplitude ratio analysis: at the key fluctuation point, combined with the data quality factor (such as light stability) of the auxiliary scanning area, calculate the phase difference and amplitude ratio of other parameters and the target parameter (such as power): phase difference: fluctuation time difference (such as reflectivity change earlier than power decline); amplitude ratio: fluctuation intensity ratio (such as the correlation between curvature change amplitude and power loss).
[0054] In this embodiment, the correlation correction factor is generated: Granger causality test: verify whether the feature parameters have statistical causality (such as whether reflectivity change causes power fluctuation). Wavelet decomposition: analyze the correlation strength of different frequency bands (such as high-frequency noise, low-frequency trend), and exclude environmental interference.
[0055] In this embodiment, the final influence weight is obtained, the parameter weight with high causality and consistent across frequency bands is strengthened, and accidental correlation is weakened.
[0056] The beneficial effects of the above technical solutions are: through joint analysis of time sequence and frequency domain, the accuracy of weight allocation is improved, and the digital twin model reflects the true physical correlation.
[0057] In some embodiments of the present application, in generating the model correction coefficient at the scanning time based on the point cloud feature parameter deviation degree at each scanning time and the current time, the final influence weight and the density compensation coefficient, it includes: Based on the Euclidean distance change rate and Hausdorff distance of the point cloud feature parameter, calculate the space-time offset; Combined with the final influence weight and the density compensation coefficient, construct a Gaussian mixture model to output the correction coefficient at the scanning time; Use Kalman filtering to perform time sequence smoothing processing on the correction coefficient.
[0058] In this embodiment, the spatio-temporal offset is calculated: the spatial deformation speed of the point cloud at adjacent scanning time points is quantified by the Euclidean distance change rate (such as the component displacement rate), the overall shape difference of the point cloud is evaluated by using the Hausdorff distance (to detect structural changes), and the two are combined to form the spatio-temporal offset, which reflects the degree of change of the physical state of the component.
[0059] In this embodiment, the Gaussian mixture model is constructed: the final influence weight (feature importance) is used as the mixing weight, the density compensation coefficient is used to adjust the variance parameter of each Gaussian distribution, the spatio-temporal offset is input into the model, and a probabilistic correction coefficient (in the range of 0-1) is output.
[0060] In this embodiment, the time series smoothing processing is performed: Kalman filtering is used to eliminate the coefficient jitter caused by measurement noise, the next time correction coefficient is predicted based on the system dynamics model, and the optimal estimation of the coefficient is realized through observation update to ensure the time series continuity.
[0061] The beneficial effects of the above technical solutions are: stable and reliable correction coefficients are generated, which are used to dynamically adjust the prediction output of the digital twin model, and the tracking ability of the model to real physical changes is improved. The whole process realizes the closed-loop optimization from spatial deformation detection to time series model correction.
[0062] In some embodiments of the present application, when the performance data and the model correction coefficient at all scanning time points within a preset time period are used to construct the digital twin model of the photovoltaic power station at the current time, the method comprises: constructing an initial digital twin model of the photovoltaic power station based on the performance data; acquiring initial model parameters of the initial digital twin model of the photovoltaic power station, and correcting the initial model parameters based on the model correction coefficient to obtain corrected initial model parameters; obtaining the digital twin model of the photovoltaic power station according to the corrected initial model parameters.
[0063] In some embodiments of the present application, when the initial model parameters of the initial digital twin model of the photovoltaic power station are acquired, and the initial model parameters are corrected based on the model correction coefficient to obtain corrected initial model parameters, the method comprises: pre-setting a first preset model correction coefficient and a second preset model correction coefficient; pre-setting a first preset adjustment coefficient, a second preset adjustment coefficient, and a third preset adjustment coefficient; when the model correction coefficient is less than the first preset model correction coefficient, the product value of the first preset adjustment coefficient and the initial model parameter is calculated as the corrected initial model parameter; When the model correction coefficient is greater than or equal to the first preset model correction coefficient and less than the second preset model correction coefficient, a product value of the second preset adjustment coefficient and the initial model parameter is calculated as the corrected initial model parameter. When the model correction coefficient is greater than or equal to the second preset model correction coefficient, a product value of the third preset adjustment coefficient and the initial model parameter is calculated as the corrected initial model parameter.
[0064] In the embodiment, the initial model parameter includes a convolution kernel size, a number of attention heads, and a hidden layer dimension.
[0065] In the embodiment, the first preset model correction coefficient is less than the second preset model correction coefficient, the first preset model correction coefficient is preferably 4, and the second preset model correction coefficient is preferably 7.
[0066] In the embodiment, the first preset adjustment coefficient is less than the second preset adjustment coefficient which is less than the third preset adjustment coefficient, the first preset adjustment coefficient is preferably 0.85, the second preset adjustment coefficient is preferably 1.05, and the third preset adjustment coefficient is preferably 1.25.
[0067] The technical scheme has the beneficial effects that the initial model parameter of the initial photovoltaic power station digital twin model is collected, and the initial model parameter is corrected based on the model correction coefficient to obtain the corrected initial model parameter, thereby further ensuring the accuracy and precision of three-dimensional digital twin modeling of the photovoltaic power station, realizing device state visualization and intelligent analysis, and providing high-precision and high-efficiency digital support for intelligent operation and maintenance of the power station.
[0068] In some embodiments of the present application, after the photovoltaic power station digital twin model at the current time is constructed based on the performance data at all scanning times in the preset time period and the model correction coefficient, the method further includes: When the point cloud data addition exceeds a threshold or the density matching degree is abnormal, the model incremental training is triggered; Fault early warning is performed based on a residual error between an output of the photovoltaic power station digital twin model and actual data, and scanning quality abnormality alarm is associated.
[0069] In the embodiment, the model incremental training triggering condition is that the point cloud data addition exceeds a threshold: when the scanning data amount significantly increases (for example, daily newly added scanning area > 20%), the model incremental learning is triggered to avoid that the historical data is out of date. The density matching degree is abnormal: if the point cloud density distribution and the reference atlas difference continuously exceed a threshold (for example, matching degree < 60%), it is determined that the scanning quality is abnormal, and the model needs to be retrained to adapt to new data features.
[0070] In this embodiment, the fault early warning mechanism is residual error analysis: the deviation of the predicted value (such as power generation) of the digital twin model from the actual monitoring data is calculated to obtain the standardized residual error. Short-term high residual error: possible instantaneous occlusion (such as bird shadow), triggering low-level warning. Persistent high residual error: indicating component degradation or failure (such as hot spot), triggering high-level warning. Scan quality association: when the residual error anomaly coincides with the low-density region in space and time, the scan device maintenance suggestion (such as lens cleaning or calibration) is pushed synchronously.
[0071] The beneficial effects of the above technical solutions are: the present application realizes early fault detection and accurate cause positioning through dynamic updating and multi-source data cross-validation.
[0072] In order to further illustrate the technical idea of the present application, the technical solutions of the present application will be described in combination with specific application scenarios.
[0073] Correspondingly, as shown in Figure 2 The present application also provides a photovoltaic power station three-dimensional digital twin modeling system based on point cloud scanning, which comprises: A point cloud scanning module is configured to acquire point cloud data and corresponding performance data at each scanning time within a preset time period before the current time through a three-dimensional scanning device during operation of the photovoltaic power station, and simultaneously collect point cloud data density characteristic values of the scanning scene; An abnormality matching module is configured to generate a historical point cloud missing distribution map and a current point cloud missing distribution map when the point cloud data density characteristic values are lower than a first density threshold, and compare the matching degrees with a preset reference missing distribution map, and trigger a data collection abnormality warning according to the matching result; An index calculation module is configured to calculate dynamic fluctuation indexes of each point cloud characteristic parameter at the scanning time according to local changes of each point cloud characteristic parameter and performance data at each scanning time, and combine dynamic compensation results of the point cloud data density characteristic values on the scanning quality; A model construction module is configured to determine the influence weight of the point cloud characteristic parameter on the performance data, generate a model correction coefficient at the scanning time based on the influence weight and the density compensation coefficient of each scanning time and the current time, and construct a digital twin model of the photovoltaic power station at the current time based on the performance data and the model correction coefficient of all scanning times within the preset time period.
[0074] In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.
[0075] Although the present application has been described with reference to the above embodiments, various modifications can be made to the application and equivalents thereof without departing from the scope of the application. In particular, features of the disclosed embodiments can be used in any combination without departing from the scope of the application, and the description of the various embodiments does not imply that the combinations of features are not combinable unless the description states that a combination is not possible. The description of the various embodiments is not meant to limit the application but merely to provide examples of the application.
[0076] It is to be understood that the above description is merely a preferred example of the application and is not intended to limit the application. Modifications can be made to the embodiments described above and equivalents thereto without departing from the scope of the application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the application shall be included in the scope of the protection of the application.
Claims
1. A three-dimensional digital twin modeling method for photovoltaic power stations based on point cloud scanning, characterized in that: include: During the operation of the photovoltaic power station, the point cloud data and corresponding performance data at each scanning moment in the preset time period before the current moment are obtained through the 3D scanning equipment, and the density characteristic value of the point cloud data of the scanning scene is collected at the same time; When the point cloud data density characteristic value is lower than the first density threshold, a historical point cloud missing distribution map and a current point cloud missing distribution map are generated, and a matching degree comparison is performed with the preset reference missing distribution map. A data collection abnormality alarm is triggered based on the matching result. Based on the local changes of each point cloud feature parameter and performance data at each scanning moment, combined with the dynamic compensation results of the point cloud data density feature value on the scanning quality, the dynamic fluctuation index of each point cloud feature parameter at the scanning moment is calculated; Determine the influence weight of point cloud feature parameters on performance data, generate the model correction coefficient at the scanning moment based on the influence weight and density compensation coefficient of each scanning moment and the current moment, and construct the digital twin model of the photovoltaic power station at the current moment based on the performance data and model correction coefficient of all scanning moments in the preset time period.
2. The photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to claim 1 is characterized in that: Based on the local changes of each point cloud feature parameter and performance data at each scanning moment, combined with the dynamic compensation result of the point cloud data density feature value on the scanning quality, the dynamic fluctuation index of each point cloud feature parameter at the scanning moment is calculated, including: For any scanning moment, the geometric features and reflectivity features in the point cloud data at any scanning moment are extracted as feature parameters; When the illumination intensity and component temperature of the scanning scene exceed a threshold, a compensation scanning configuration is initiated and a scanning stability period of a transition scanning area is calculated; A time window is set with any scanning moment as the center, and the standard deviation, coefficient of variation and local entropy of each characteristic parameter and performance data in the time window are calculated in combination with the density compensation coefficient to generate a dynamic fluctuation index.
3. The photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to claim 1 is characterized in that: When determining the weight of the influence of point cloud feature parameters on performance data, include: In a preset time period, the differences between the point cloud feature parameters and the performance data in terms of dynamic fluctuation index and density compensation coefficient are integrated to determine the initial influence weight of each point cloud feature parameter on the performance data; According to the dynamic fluctuation correlation between each point cloud feature parameter and its initial influence weight, the final influence weight of each parameter on the performance data is calculated.
4. The photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to claim 3 is characterized in that: When fusing the differences between the point cloud feature parameters and the performance data in terms of dynamic fluctuation index and density compensation coefficient, and determining the initial influence weight of each point cloud feature parameter on the performance data, the following are included: Calculate the Pearson correlation coefficient and time-delay mutual information between each point cloud feature parameter and performance data within a preset time period; Quantify the fluctuation time series deviation of characteristic parameters and performance data through dynamic time warping algorithm; The Pearson correlation coefficient, time-lagged mutual information, fluctuation time series bias and density compensation coefficient are weighted and fused, and the initial influence weight is obtained after normalization.
5. The photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to claim 3 is characterized in that: When calculating the final influence weight of each parameter on the performance data based on the dynamic fluctuation correlation between each point cloud feature parameter and its initial influence weight, it includes: The first K extreme value points of the dynamic fluctuation index of each characteristic parameter are selected as key fluctuation points; At the critical fluctuation point, the fluctuation phase difference and amplitude ratio between other parameters and the target parameter are calculated in combination with the data quality influencing factors of the auxiliary scanning area; Generate correlation correction factors based on cross-frequency band correlation analysis using Granger causality test and wavelet decomposition; The product of the association correction factor and the initial impact weight is taken as the final impact weight.
6. The photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to claim 1 is characterized in that: When generating the model correction coefficient at the scanning moment based on the deviation of the point cloud feature parameters at each scanning moment and the current moment, the final influence weight and density compensation coefficient include: The spatiotemporal offset is calculated based on the Euclidean distance change rate and Hausdorff distance of the point cloud feature parameters; The Gaussian mixture model is constructed by combining the final influence weight and density compensation coefficient to output the correction coefficient at the scanning moment; Kalman filtering is used to perform time series smoothing on the correction coefficients.
7. The photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to claim 1 is characterized in that: When building a digital twin model of a photovoltaic power station at the current moment based on the performance data and model correction coefficients at all scanning moments within a preset time period, the following steps are included: Building an initial digital twin model of the photovoltaic power station based on the performance data; Collecting initial model parameters of the initial photovoltaic power station digital twin model, and correcting the initial model parameters based on the model correction coefficient to obtain corrected initial model parameters; A digital twin model of a photovoltaic power station is obtained according to the modified initial model parameters.
8. The photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to claim 7 is characterized in that: When collecting initial model parameters of the initial photovoltaic power station digital twin model and correcting the initial model parameters based on the model correction coefficient to obtain the corrected initial model parameters, the method includes: presetting a first preset model correction coefficient and a second preset model correction coefficient; Presetting a first preset adjustment coefficient, a second preset adjustment coefficient, and a third preset adjustment coefficient; When the model correction coefficient is less than the first preset model correction coefficient, the product of the first preset adjustment coefficient and the initial model parameter is calculated as the corrected initial model parameter; When the model correction coefficient is greater than or equal to the first preset model correction coefficient and less than the second preset model correction coefficient, a product value of the second preset adjustment coefficient and the initial model parameter is calculated as the correction initial model parameter; When the model correction coefficient is greater than or equal to the second preset model correction coefficient, the product value of the third preset adjustment coefficient and the initial model parameter is calculated as the corrected initial model parameter.
9. The photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to claim 1 is characterized in that: After building the digital twin model of the PV power station at the current moment based on the performance data and model correction coefficients at all scan times within the preset time period, the following also applies: When the amount of new point cloud data exceeds the threshold or the density matching is abnormal, the model incremental training is triggered; Fault warnings are issued based on the residuals between the output of the digital twin model of the photovoltaic power station and the actual data, and associated with abnormal scanning quality alarms.
10. A photovoltaic power station three-dimensional digital twin modeling system based on point cloud scanning, applied to the photovoltaic power station three-dimensional digital twin modeling method based on point cloud scanning according to any one of claims 1 to 9, characterized in that: include: The point cloud scanning module is used to obtain point cloud data and corresponding performance data at each scanning moment within a preset time period before the current moment through a three-dimensional scanning device during the operation of the photovoltaic power station, and at the same time collect the point cloud data density characteristic value of the scanning scene; Anomaly matching module, which is used to generate a historical point cloud missing distribution map and a current point cloud missing distribution map when the point cloud data density characteristic value is lower than a first density threshold, and compare the matching degree with the preset reference missing distribution map, and trigger a data collection abnormality alarm based on the matching result; The index calculation module is used to calculate the dynamic fluctuation index of each point cloud feature parameter at the scanning moment based on the local changes of each point cloud feature parameter and performance data at each scanning moment, combined with the dynamic compensation result of the point cloud data density feature value on the scanning quality; The model construction module is used to determine the influence weight of point cloud feature parameters on performance data, generate the model correction coefficient at the scanning moment based on the influence weight and density compensation coefficient of each scanning moment and the current moment, and construct the digital twin model of the photovoltaic power station at the current moment based on the performance data and model correction coefficient of all scanning moments within the preset time period.