Distributed photovoltaic generating capacity real-time monitoring and multi-dimensional operation index linkage analysis system and method

Abstract

The invention discloses a distributed photovoltaic generating capacity real-time monitoring and multi-dimensional operation index linkage analysis system and method. The system comprises a current data acquisition module which is used for acquiring distributed photovoltaic generating capacity real-time monitoring data; the core feature calculation module is used for calculating component-level core features, multi-scale factors and principal component response intensity; the coupling weight and dynamic coefficient calculation module is used for generating a coupling weight and a dynamic coefficient; the space-time nonlinear coupling matrix generation module is used for generating a space-time nonlinear coupling matrix; and the generating capacity real-time prediction and deviation analysis module is used for performing generating capacity real-time prediction and deviation analysis. According to the method, a linkage mechanism of real-time working conditions and parameter adjustment is set in the links of core feature calculation and coupling weight and dynamic coefficient generation, so that complex working conditions such as illumination fluctuation, temperature change and local shielding of the photovoltaic power station can be used in the scheme.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and more particularly to a system and method for real-time monitoring of power generation of distributed photovoltaic and linkage analysis of multi-dimensional operation indicators. Background Art

[0002] With the global energy structure transformation towards clean and renewable energy, distributed photovoltaic power generation has been widely promoted and applied due to its advantages such as flexible installation and nearby consumption. However, in the actual operation process of distributed photovoltaic power stations, there are problems such as untimely and inaccurate power generation monitoring, and single operation index analysis. Currently, most of the existing distributed photovoltaic monitoring systems can only achieve simple collection and display of power generation data, unable to capture the dynamic changes of power generation in real time. In terms of operation index analysis, they often only focus on a few indicators such as the temperature and light intensity of photovoltaic modules, lacking comprehensive consideration of multi-dimensional operation indicators such as the operating state of inverters, grid voltage frequency, and environmental temperature and humidity, and are even less able to achieve linkage analysis between power generation and these multi-dimensional operation indicators. This makes it difficult for operation and maintenance personnel to accurately judge the reasons for power generation fluctuations, unable to timely detect potential faults in the operation of photovoltaic power stations, thus affecting the power generation efficiency and service life of photovoltaic power stations and increasing operation and maintenance costs. Therefore, there is an urgent need for a system that can achieve real-time monitoring of power generation and conduct linkage analysis with multi-dimensional operation indicators to solve the deficiencies in the existing technology and improve the operation and management level of distributed photovoltaic power stations. Summary of the Invention

[0003] The purpose of the present invention is to provide a system for real-time monitoring of power generation of distributed photovoltaic and linkage analysis of multi-dimensional operation indicators to at least solve one of the above technical problems.

[0004] In one aspect of the present invention, there is provided a system for real-time monitoring of power generation of distributed photovoltaic and linkage analysis of multi-dimensional operation indicators, and the system for real-time monitoring of power generation of distributed photovoltaic and linkage analysis of multi-dimensional operation indicators includes: A current data acquisition module, which is used to acquire real-time monitoring data of the power generation of distributed photovoltaic; A core feature calculation module, which is used to calculate component-level core features, multi-scale factors, and principal component response intensities according to the real-time monitoring data of the power generation of distributed photovoltaic; A coupling weight and dynamic coefficient calculation module, which is used to generate coupling weights and dynamic coefficients according to component-level core features, multi-scale factors, and principal component response intensities; A spatiotemporal nonlinear coupling matrix generation module is used to generate a spatiotemporal nonlinear coupling matrix based on coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity. A real-time power generation prediction and deviation analysis module is used to perform real-time power generation prediction and deviation analysis based on a spatiotemporal nonlinear coupling matrix.

[0005] Optionally, the real-time monitoring and multi-dimensional operational indicator linkage analysis system for distributed photovoltaic power generation further includes: A multi-dimensional indicator anomaly diagnosis module is used to perform tiered investigation when the analysis results of the real-time power generation prediction and deviation analysis module exceed a preset safety threshold.

[0006] Optionally, the real-time power generation monitoring data includes component-level data, inverter-level data, grid-level data, environmental-level data, and power generation data.

[0007] Optionally, the component-level core features are calculated through the following steps: The neighborhood influence coefficient is calculated based on real-time power generation monitoring data. Calculate the time-series decay coefficient based on real-time power generation monitoring data; Spatial illumination cross characteristics are calculated based on real-time power generation monitoring data and neighborhood influence coefficient. The time-temperature crossover characteristics are calculated based on real-time power generation monitoring data and time-series decay coefficient. Spatiotemporal power crossover characteristics are calculated based on real-time power generation monitoring data.

[0008] Optionally, the multi-scale factor is calculated through the following steps: The component-level, string-level, and subarray-level coefficients of variation are calculated using the neighborhood influence coefficient as the calculation object. The spatial scale factor is obtained based on the component-level and subarray-level coefficients of variation. Calculate the time scale factor based on the time decay coefficient; The spatiotemporal crossover scale factor is calculated based on the spatial scale factor, the temporal scale factor, and the spatiotemporal power crossover characteristics.

[0009] Optionally, the principal component response intensity is obtained in the following manner: Obtain the trained KPCA dimensionality reduction model; The spatial illumination cross features, temporal temperature cross features, and spatiotemporal power cross features were dimensionality reduced using a trained KPCA dimensionality reduction model to extract three principal components and obtain the principal component response intensities corresponding to the three principal components.

[0010] Optionally, generating the spatiotemporal nonlinear coupling matrix based on coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity includes: Obtain the response scale rule base; The intensity of the principal component response intensity corresponding to each principal component is graded to obtain the corresponding level of each principal component; The spatial scale factor, temporal scale factor, and spatiotemporal cross-scale factor are classified to obtain the spatial scale factor level, temporal scale factor level, and spatiotemporal cross-scale factor level. The weight formula coefficients are obtained from the response scale rule base according to the corresponding level, spatial scale factor level, temporal scale factor level, and spatiotemporal cross-scale factor level of each principal component. Based on the obtained weight formula coefficients, obtain the initial values ​​of spatial weight, time weight, and spatiotemporal cross weight respectively; The initial values ​​of spatial weights, temporal weights, and spatiotemporal cross-weights are calibrated. Obtain the trained dynamic coefficient generation model; The response intensity of each principal component, spatial scale factor, temporal scale factor, spatiotemporal cross scale factor, calibrated initial values ​​of spatial weights, calibrated initial values ​​of temporal weights, and calibrated initial values ​​of spatiotemporal cross weights are input into the trained dynamic coefficient generation model to obtain dynamic coefficients, which include spatial coefficients, temporal coefficients, and spatiotemporal cross coefficients.

[0011] Optionally, generating the spatiotemporal nonlinear coupling matrix based on coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity includes: Obtain the preset coupling function; The obtained coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity are substituted into the preset coupling function to obtain the coupling feature values ​​of a single component at a single time node. Obtain the subarray-level real-time coupling feature value sequence based on the coupling feature value of each single component at a single time node; Generate a three-dimensional coupling matrix of subarray-timescale-timenode based on the real-time coupled eigenvalue sequence of subarray level; The spatiotemporal nonlinear coupling matrix is ​​obtained by reducing the dimensionality of the subarray-timescale-timenode three-dimensional coupling matrix using the Tucker decomposition algorithm.

[0012] Optionally, the real-time prediction and deviation analysis of power generation based on the spatiotemporal nonlinear coupling matrix includes: Obtain the trained optical optimization matrix neural network; The spatiotemporal nonlinear coupling matrix is ​​input into the optical optimization matrix neural network to obtain the predicted power generation value for a future preset time period; Obtain historical power generation data for the same period; The deviation rate is calculated based on historical power generation data for the same period and the projected power generation for a future time period. The operating status is obtained based on the deviation rate; The correlation degree of deviation source tracing is obtained based on the deviation rate and the response intensity of each principal component.

[0013] This application also provides a method for real-time monitoring of power generation and multi-dimensional operational index linkage analysis of distributed photovoltaic power generation, characterized in that the method includes: Obtain real-time monitoring data on the power generation of distributed photovoltaic systems; The core characteristics, multi-scale factors, and principal component response intensity of the components are calculated based on real-time monitoring data of distributed photovoltaic power generation. Coupling weights and dynamic coefficients are generated based on component-level core features, multi-scale factors, and principal component response intensity. A spatiotemporal nonlinear coupling matrix is ​​generated based on coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity. Real-time prediction and deviation analysis of power generation are performed based on the spatiotemporal nonlinear coupling matrix.

[0014] The distributed photovoltaic (PV) power generation real-time monitoring and multi-dimensional operation index linkage analysis data system of this application incorporates a linkage mechanism for real-time operating conditions and parameter adjustments in the core feature calculation, coupling weight, and dynamic coefficient generation stages. This allows the solution to utilize complex operating conditions of PV power plants, such as light fluctuations, temperature changes, and local shading. For example, during cloudy weather with frequent light fluctuations, the time-series attenuation coefficient and dynamic coefficient can be adjusted in real time to follow the power trend, avoiding feature calculation deviations caused by fixed parameters. In scenarios where roof edge modules are easily shaded, the dynamic range adjustment of the neighborhood influence coefficient can accurately capture the impact of shading on module power, providing reliable feature data for subsequent anomaly diagnosis and efficiency optimization. Using the method of this application, the stability of analysis results under different operating conditions can be achieved, avoiding a sudden drop in accuracy or analysis failure due to changes in operating conditions, and ensuring the reliability of operation and maintenance decisions. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system according to an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0017] like Figure 1 The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system shown includes a current data acquisition module, a core feature calculation module, a coupling weight and dynamic coefficient calculation module, a spatiotemporal nonlinear coupling matrix generation module, and a power generation real-time prediction and deviation analysis module; among which, The current data acquisition module is used to acquire real-time monitoring data on the power generation of distributed photovoltaic systems. The core feature calculation module is used to calculate the component-level core features, multi-scale factors, and principal component response intensity based on real-time monitoring data of distributed photovoltaic power generation. The coupling weight and dynamic coefficient calculation module is used to generate coupling weights and dynamic coefficients based on component-level core features, multi-scale factors, and principal component response intensity. The spatiotemporal nonlinear coupling matrix generation module is used to generate a spatiotemporal nonlinear coupling matrix based on coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity. The real-time power generation prediction and deviation analysis module is used to perform real-time power generation prediction and deviation analysis based on the spatiotemporal nonlinear coupling matrix.

[0018] In this embodiment, the real-time power generation monitoring data includes component-level data, inverter-level data, grid-level data, environmental-level data, and power generation data.

[0019] In this embodiment, component-level data includes temperature data, current data, voltage data, and illumination data. Specifically, the temperature, operating current, operating voltage, and corresponding illumination intensity of all components within the target monitoring subarray are acquired through integrated sensors and distributed acquisition modules. The acquisition frequency is consistent with the component power sampling frequency (≤1 second / time). Inverter-level data includes inverter input power, inverter output power, inverter power factor, and inverter temperature. Specifically, it includes the inverter input power, output power, power factor, and inverter temperature corresponding to the target subarray. The inverter's temperature and operating status code (normal / abnormal) are read in real time via the inverter's communication interface (such as RS485 or Ethernet); grid-level data includes voltage, frequency, and harmonic data; specifically, the voltage, frequency, and harmonic content (total harmonic distortion rate and the proportion of each harmonic) at the power plant's grid-connected side and subarray grid-connected points are collected by power quality monitoring devices; environmental-level data includes temperature and humidity, wind speed, and wind direction data, which can be obtained through environmental monitoring stations; power generation-level data includes total power generation, subarray-level power generation, and module-level instantaneous power generation, etc. In this embodiment, the acquired data can be preprocessed, such as data verification, abnormal data processing, missing data filling, data standardization and other conventional processing.

[0020] In existing technologies, distributed photovoltaic power stations are usually composed of multiple subarrays. Taking a distributed photovoltaic power station with a total installed capacity of 1MW as an example, it is usually divided into 10 subarrays (numbered Z1-Z10, according to the inverter configuration habit of a 1MW power station, 10 100kW inverters correspond to 10 subarrays), and each subarray contains 400 250W photovoltaic modules. In existing technologies, the component layout parameters are usually as follows: planar roof layout, component spacing of 0.3m (refer to the minimum spacing requirements for planar layout in GB50797-2012 "Design Code for Photovoltaic Power Stations" to ensure no fixed obstruction).

[0021] In this embodiment, the component-level core features are calculated through the following steps: The neighborhood influence coefficient is calculated based on real-time power generation monitoring data. Specifically, for each module in the target subarray, the neighborhood range is determined according to the module arrangement (planar / sloping), and the average power ratio between the module and the neighboring modules is calculated to characterize the degree of influence of the module on the operating status of the adjacent modules. The time-series decay coefficient is calculated based on real-time power generation monitoring data. Specifically, linear regression is performed on the component power data using a preset time window (usually 10 minutes) to calculate the power decay trend coefficient over time, which characterizes the time-series change characteristics of component power due to factors such as temperature rise and light change. Spatial illumination cross-characteristics are calculated based on real-time power generation monitoring data and neighborhood influence coefficients; temporal temperature cross-characteristics are calculated based on real-time power generation monitoring data and temporal decay coefficients; and spatiotemporal power cross-characteristics are calculated based on real-time power generation monitoring data.

[0022] In this embodiment, the spatial illumination cross feature represents the ratio of the average illumination intensity of the neighborhood to the illumination intensity of the component itself; the temporal temperature cross feature represents the ratio of the component's real-time temperature to its average temperature over the past hour; and the spatiotemporal power cross feature represents the product of the ratios of the component's real-time power to the average power of the subarray and the power of the same period yesterday.

[0023] In this embodiment, the neighborhood range can be determined as follows: Planar layout (such as rooftop array): Take a 3×3 component matrix as the neighborhood, that is, the target component has 8 adjacent components around it (2 above, below, left, right, and diagonally). The neighborhood range parameter is stored in the configuration file and can be modified according to the actual layout of the power plant. Sloping layout (such as photovoltaic greenhouse): Since the sloping modules are only adjacent to each other and are significantly affected by sunlight shading, two modules above and below are taken as the neighborhood, for a total of four adjacent modules.

[0024] In this embodiment, the neighborhood influence coefficient is obtained using the following formula: ;in, Let P1 be the average power of the target component, and S be the neighborhood influence coefficient, typically ranging from 0.8 to 1.2. A value greater than 1.2 indicates that the neighboring components are performing better than the target component, while a value less than 0.8 indicates that the target component is negatively affected by the neighborhood (e.g., insufficient illumination due to shading by neighboring components). The average power of the neighboring components is: P1, P2, ..., Pn (where n is the number of neighboring components). The average power of the k-th neighboring component; In this embodiment, if the calculation result S > 1.5 or S < 0.5, it is judged as abnormal (possibly due to the failure of the neighboring component or the error in data acquisition). The mean of the neighborhood influence coefficient of the same type of component in the same subarray is used to replace it to ensure the validity of the feature data.

[0025] In this embodiment, the time-series decay coefficient is analyzed by linear regression to determine the trend of component power change within a preset time window, thereby quantifying the degree of power decay over time (such as the slow decrease in power due to temperature rise).

[0026] First, a preset time window parameter needs to be called (usually 10 minutes, which can be adjusted according to needs, with a minimum of 5 minutes and a maximum of 15 minutes), to collect real-time power data of the target component at 1 minute / time, obtaining a total of 10 power values: P t1 ,P t2 ,...,P t10The timestamps are t1, t2, ..., t 10 (1 minute interval); Using timestamps as independent variables (converted to minutes, e.g., t1=0 minutes, t2=1 minute, ..., t...), 10 =9 minutes), with power as the dependent variable, linear regression is performed to obtain the regression equation: P=kt+b, where k is the slope (unit: kW / minute), which characterizes the rate of change of power over time (k<0 indicates power decay, k>0 indicates power increase).

[0027] First, calculate the average power within the time window: ; The average power of the target component within a preset time window; Let be the real-time power of the target component during the i-th data acquisition. Then substitute the values ​​into the following formula to calculate the timing decay coefficient: Where T is the time-series decay coefficient, with a value range of 0.8-1.2. >1 indicates that the power increases with time (such as increased light intensity), and <1 indicates that the power decreases (such as increased temperature). If the absolute value of the slope k is greater than 0.1 kW / min (i.e., the power change exceeds 1 kW within 10 minutes), it is judged as an abnormal trend (possibly due to component failure or extreme weather). Data is collected again for calculation. If there are 3 consecutive abnormalities, a power fluctuation warning is triggered.

[0028] In this embodiment, the calculation of spatial illumination cross characteristics based on real-time power generation monitoring data and neighborhood influence coefficient includes: Real-time illumination intensity of target components (Unit: W / m²), Real-time illuminance of neighboring components I1, I2, ..., I n The average illuminance of the neighborhood is calculated using the following formula: ; The spatial illumination cross-feature is calculated using the following formula: Where S is the neighborhood influence coefficient; It is a spatial illumination intersection feature; The average illumination intensity of the neighboring components; The real-time illumination intensity of the target component; Standardization processing: Based on the preset maximum light intensity (usually 2000W / ㎡) and minimum value (0W / ㎡), Map to the interval [0,1].

[0029] In this embodiment, the calculation of time-temperature crossover characteristics based on real-time power generation monitoring data and time-series decay coefficient includes: Collect real-time temperature of target component (Unit: °C) Average temperature over the past hour (Calculate the average temperature data from the past 60 minutes retrieved from the real-time database). Substitute into the following formula: Where T is the time-series decay coefficient, For time-temperature crossover characteristics, Real-time temperature of the target component, This represents the average temperature over the past hour. Standardization process: Based on the preset maximum component temperature (usually 60℃) and minimum temperature (-20℃), Map to the interval [0,1].

[0030] In this embodiment, calculating the spatiotemporal power crossover characteristics based on real-time power generation monitoring data includes: Obtain the real-time average power of the subarray (Average power of all components within the subarray), Power of the target component at the same time yesterday (Retrieve power data at the same time from the historical database); Substitute into the following formula: ;in, Spatiotemporal power crossover characteristics; The average power of the target component; This represents the real-time average power of the subarray (the average real-time power of all components within the subarray). The target component's power output was the same as yesterday.

[0031] Standardization processing: Based on the preset maximum power ratio (usually 1.5) and minimum power ratio (0.5), the power ratio is then... Map to the interval [0,1].

[0032] In this embodiment, the multi-scale factor is calculated through the following steps: The component-level, string-level, and subarray-level coefficients of variation are calculated using the neighborhood influence coefficient as the calculation object. In this embodiment, the component-level coefficient of variation is obtained in the following way: Select the S values ​​of all components (usually 40-50 blocks) within any given string, and calculate the coefficient of variation using the following formula: ;in, The standard deviation of the component-level S-value. The mean of the component-level S-values; The string-level coefficient of variation is obtained as follows: Select the average S-value (mean S-value of components within each string) of all strings in any subarray (usually 10-12 strings), and calculate the string-level coefficient of variation using the following formula: ;in, The standard deviation of the average neighborhood influence coefficient at the string level; This represents the mean of the average neighborhood influence coefficient at the string level. Subarray-level variation coefficients are obtained as follows: Select the average S-value (mean S-value of strings within each subarray) of all subarrays (usually 10-20) and calculate the subarray-level coefficient of variation using the following formula: ;in, is the standard deviation of the average neighborhood influence coefficient at the subarray level; This represents the mean of the average neighborhood influence coefficient at the subarray level. The spatial scale factor is obtained based on the component-level and subarray-level coefficients of variation. In this embodiment, the spatial scale factor is obtained based on the component-level coefficient of variation and the subarray-level coefficient of variation using the following formulas: ;in, Spatial scale factor; The coefficient of variation at the subarray level; This represents the component-level coefficient of variation.

[0033] Calculate the time scale factor based on the time decay coefficient; In this embodiment, calculating the time scale factor based on the time decay coefficient includes: By calculating the characteristic fluctuation ratios at three time scales—10 seconds, 1 minute, and 10 minutes—the severity of characteristic fluctuations at different time scales can be quantified (e.g., large fluctuations at the 10-second level may be due to rapid cloud movement). First, calculate the volatility at each level: Taking the time-series decay coefficient T as the calculation object: 10-second level: Collect T values ​​within 10 seconds at 1 second / time (10 data points in total), and calculate the volatility using the following formula: ;in, The fluctuation of the timing decay coefficient is on the order of 10 seconds. The time-series attenuation coefficient for the (i+1)th acquisition; Let be the timing attenuation coefficient for the i-th acquisition; 1-minute level: Collect T values ​​within 1 minute at 10-second intervals (6 data points in total), and calculate volatility using the following formula: ;in, The fluctuation of the time-series decay coefficient is measured in 1-minute increments. The time-series attenuation coefficient for the (i+1)th acquisition; Let be the timing attenuation coefficient for the i-th acquisition; 10-minute level: Collect T values ​​within 10 minutes at 1 minute intervals (10 data points in total), and calculate volatility using the following formula: ;in, The fluctuation of the time-series decay coefficient is measured in 10-minute timeframes. The time-series attenuation coefficient for the (i+1)th acquisition; Let be the timing decay coefficient for the i-th acquisition.

[0034] The time scale factor is obtained using the following formula: ;in, This is the time scale factor.

[0035] The value is usually between 1.0 and 5.0. A value greater than 1.5 indicates drastic fluctuations on a short time scale (such as instantaneous occlusion), while a value less than 0.8 indicates stable temporal characteristics.

[0036] The spatiotemporal crossover scale factor is calculated based on the spatial scale factor, the temporal scale factor, and the spatiotemporal power crossover characteristics.

[0037] In this embodiment, the spatiotemporal cross-scale factor is calculated using the following formula: ;in, It is the spatiotemporal cross-scale factor.

[0038] In this embodiment, the above three types of intersection features (( All of these need to be standardized. Standardization is a routine technical method and will not be elaborated on here.

[0039] In this embodiment, the principal component response intensity is obtained in the following manner: Obtain the trained KPCA dimensionality reduction model; The spatial illumination cross features, temporal temperature cross features, and spatiotemporal power cross features were dimensionality reduced using a trained KPCA dimensionality reduction model to extract three principal components and obtain the principal component response intensities corresponding to the three principal components.

[0040] Obtain the standardized three types of cross features: This forms the feature vector: .

[0041] Call the preset KPCA parameters: Kernel function: RBF kernel (radial basis function), parameter gamma=0.1 (determined through training optimization); Principal component vectors: V1, V2, V3 (vectors corresponding to the three principal components); Mean vector: ; Calculation process: Feature centralization: ; Kernel function mapping: ; Principal component projection: Calculate the projection values ​​of the eigenvectors onto the three principal components, i.e., the principal component response intensities. Principal Component 1 Response Intensity (Spatial Illumination Dominant): ; Principal Component 2 Response Intensity (Time and Temperature Dominant): ; Principal Component 3 Response Intensity (Spatiotemporal Power Dominant): . Step 3: Response Intensity Grading Call the preset grading threshold during the training phase: high: ; middle: ; Low: ; Output the classification results of each principal component (e.g., PC1 high, PC2 medium, PC3 low) for subsequent response scale rule base matching.

[0042] In this embodiment, the above-mentioned hierarchical classification is exemplified as follows: Principal component response intensity classification: Call preset thresholds (e.g., PC high > 0.7, medium 0.3-0.7, low < 0.3) to classify the real-time calculated PC1 (spatial illumination dominant), PC2 (temporal temperature dominant), and PC3 (spatiotemporal power dominant) into categories and generate labels (e.g., PC1 high, PC2 medium, PC3 low). Example: If PC1=0.78, PC2=0.42, and PC3=0.25, then the classification labels are PC1 High, PC2 Medium, and PC3 Low. Multiscale factor classification: Spatial scale factor F S -scale): Preset thresholds: high > 1.2, medium 0.8-1.2, low < 0.8. Labels are generated after grading (e.g., F). S high); Time scale factor (F) T -scale): Preset thresholds: high > 1.2, medium 0.8-1.2, low < 0.8. Labels are generated after grading (e.g., F). T high); Spatiotemporal cross-scale factor (F ST -scale): No separate grading is required; its value is already incorporated into the weighting formula.

[0043] Through the above classification, five classification tags can be obtained for subsequent rule base scenario matching.

[0044] In this embodiment, generating the spatiotemporal nonlinear coupling matrix based on coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity includes: Obtain the response scale rule base; The intensity of the principal component response intensity corresponding to each principal component is graded to obtain the corresponding level of each principal component; The spatial scale factor, temporal scale factor, and spatiotemporal cross-scale factor are classified to obtain the spatial scale factor level, temporal scale factor level, and spatiotemporal cross-scale factor level. The weight formula coefficients are obtained from the response scale rule base according to the corresponding level, spatial scale factor level, temporal scale factor level, and spatiotemporal cross-scale factor level of each principal component. Based on the obtained weight formula coefficients, obtain the initial values ​​of spatial weight, time weight, and spatiotemporal cross weight respectively; The initial values ​​of spatial weights, temporal weights, and spatiotemporal cross-weights are calibrated. Obtain the trained dynamic coefficient generation model; The response intensity of each principal component, spatial scale factor, temporal scale factor, spatiotemporal cross scale factor, calibrated initial values ​​of spatial weights, calibrated initial values ​​of temporal weights, and calibrated initial values ​​of spatiotemporal cross weights are input into the trained dynamic coefficient generation model to obtain dynamic coefficients, which include spatial coefficients, temporal coefficients, and spatiotemporal cross coefficients.

[0045] In this embodiment, the response scale rule base is a preset rule base, composed of 3 principal component response intensity levels (principal component 1 / principal component 2 / principal component 3, corresponding to spatial illumination features, temporal temperature features, and spatiotemporal power features, respectively) × 3 spatial scale factor levels (high H / medium M / low L) × 3 temporal scale factor levels (high H / medium M / low L). Each combination corresponds to a specific formula. The rule base uses principal component hierarchical labels + multi-scale factor hierarchical labels as indexes, corresponding to different combination scenarios. Each scenario stores the corresponding weight formula coefficients A / B / C / D / E / F (a total of 6 coefficients, corresponding to the calculation parameters of spatial / temporal / spatiotemporal cross weights).

[0046] The formula is set up as follows: The weighting formulas for all scenarios are based on a unified framework of spatial feature terms + temporal feature terms + spatiotemporal cross terms. The formulas are adapted to different scenarios by adjusting the coefficients A / B / C / D / E / F in the formulas. The formula framework is as follows: Spatial weight initial value ; Initial value of time weight ; initial values ​​of spatiotemporal cross weights .

[0047] A / B / C / D / E / F are scene adaptation coefficients (different values ​​for different scenes, ranging from 0.05 to 0.5).

[0048] The value of coefficient AF follows the principle of prioritizing dominant features: If the principal component 1 (spatial illumination characteristics) is of a high level, then A (spatial factor coefficient) takes a larger value (0.3-0.5) to strengthen the influence of spatial characteristics; If the principal component 2 (time-temperature characteristics) is at a high level, then C (time factor coefficient) takes a larger value (0.3-0.5) to strengthen the influence of time characteristics; If the spatial scale factor or temporal scale factor is of a high level, the corresponding coefficient (A / C) increases synchronously, reflecting the significance of scale differences.

[0049] The following example illustrates a typical scenario's weighting formula: Category 1: High principal component 1 (dominated by spatial illumination characteristics) → enhanced A / C1 (spatial correlation coefficient); Scenario 1: High principal component 1 + high spatial scale (H) + high temporal scale (H) Coefficient settings: A=0.3 (high weight for spatial factors), B=0.2 (auxiliary for spatial illumination characteristics), C=0.2 (medium weight for temporal factors), D=0.1 (auxiliary for temporal temperature characteristics), E=0.2 (medium weight for spatiotemporal cross-factors), F=0.1 (auxiliary for spatiotemporal power characteristics). Scenario 2: Principal Component 1 is high + spatial scale is high (H) + temporal scale is medium (M) Coefficient adjustment: The time scale is reduced, C decreases from 0.2 to 0.15, and the other coefficients remain unchanged.

[0050] Category 2: High principal component 2 (time-temperature characteristics dominate → enhanced C / D (time correlation coefficient) Scenario 1: High principal component 2 + high spatial scale (H) + high temporal scale (H) Coefficient settings: C=0.3 (high weight of time factor), D=0.2 (assistance of time and temperature features), A=0.2 (medium weight of spatial factor), B=0.1 (assistance of spatial illumination features), E=0.2 (medium weight of spatiotemporal cross-factor), F=0.1 (assistance of spatiotemporal power features).

[0051] Category 3: Principal component 3 high (spatiotemporal power characteristics dominate) → enhanced E / F (spatiotemporal cross-correlation coefficient); Scenario 1: Principal component 3 high + spatial scale high (H) + temporal scale high (H) Coefficient settings: E=0.3 (high weight for spatiotemporal cross-factor), F=0.2 (auxiliary for spatiotemporal power characteristics), A=0.15 (low weight for spatial factors), B=0.1 (auxiliary for spatial illumination characteristics), C=0.15 (low weight for temporal factors), D=0.1 (auxiliary for temporal temperature characteristics). Category 4: Principal component mixture level (not a single dominant component → balance coefficient) Scenario 1: Principal Component 1 high + Principal Component 2 medium + Principal Component 3 low + Spatial M + Temporal H Coefficient settings: A=0.25 (high principal component 1), C=0.2 (medium principal component 2 + time H), E=0.15 (low principal component 3), B / D / F=0.1 (auxiliary features).

[0052] Understandably, the coefficients can be set as needed.

[0053] In this embodiment, the weight formula coefficients obtained from the response scale rule base based on the corresponding level, spatial scale factor level, temporal scale factor level, and spatiotemporal cross-scale factor level of each principal component include: Principal component classification labels (PC1 / PC2 / PC3) and multi-scale factor classification labels (F) S / F T Combining these elements creates a scene index (e.g., PC1 High - PC2 Medium - PC3 Low - F). S High-F T high); Coefficient retrieval: Match the corresponding coefficient set in the rule base based on the scene index. For example: if the scene is PC1 High - PC2 Medium - PC3 Low - F S High-F T High, retrieve coefficients A=0.3, B=0.2, C=0.2, D=0.1, E=0.2, F=0.1 (coefficient values ​​are preset to ensure optimal weight calculation accuracy in this scenario); Using the above framework formula, the initial values ​​of the spatial weights can be obtained respectively. , as well as .

[0054] In this embodiment, calibrating the initial values ​​of spatial weights, temporal weights, and spatiotemporal crossover weights includes: Calculate the photovoltaic characteristic calibration factor (K) PV ): Call standard test condition parameters (lighting I) std =1000W / ㎡, temperature T std =25℃), substituting into the formula: ; (I real For real-time light intensity, T real (for the real-time temperature of the components). Initial weight standardization: Calculate the initial weight sum Standardize the initial weights (to ensure the sum is 1).

[0055] Post-calibration weight calculation: Introduce a calibration index (preset, usually 0.3, to ensure a reasonable calibration range), formula: ; Weighted double standardization: Calculate the total weight after calibration If the sum is not equal to 1, scale it proportionally to a sum of 1 to obtain the calibrated coupling weights (W) with a sum of 1. S / W T / W ST ) In this embodiment, the dynamic coefficient generation model is the GBT model; The input to the Dynamic Coefficient Generation Model (GBT) consists of nine quantitative features, as follows: Principal component response intensities (3): PC1, PC2, PC3 (standardized values); Multiscale factors (3): F S -scale, F T -scale, F ST-scale (raw calculated value, which already contains scale difference information and does not require secondary standardization); Coupling weights (3): W S / W T / W ST (Value after calibration).

[0056] According to the model's preset input order, such as: PC1→PC2→PC3→F S -scale→F T -scale→F ST -scale→W S →W T →W ST The nine feature values ​​are used to form an input vector, which is then input into the trained GBT model to obtain three continuous dynamic coefficients (K). S Spatial coefficient, K T Time coefficient, K ST Spatiotemporal crossover coefficient).

[0057] In this embodiment, the GBT model in the training phase learns the mapping relationship between input features and dynamic coefficients by integrating multiple decision trees.

[0058] In this embodiment, generating the spatiotemporal nonlinear coupling matrix based on coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity includes: Obtain the preset coupling function; The obtained coupling weights, dynamic coefficients, component-level core features, multi-scale factors, and principal component response intensity are substituted into the preset coupling function to obtain the coupling feature values ​​of a single component at a single time node. Obtain the subarray-level real-time coupling feature value sequence based on the coupling feature value of each single component at a single time node; Generate a three-dimensional coupling matrix of subarray-timescale-timenode based on the real-time coupled eigenvalue sequence of subarray level; The spatiotemporal nonlinear coupling matrix is ​​obtained by reducing the dimensionality of the subarray-timescale-timenode three-dimensional coupling matrix through the Tucker decomposition algorithm.

[0059] In this embodiment, the coupling function formula is as follows: ;in, S is the coupling characteristic value of a single component at a single time node; S is the component's neighborhood influence coefficient; T is the component's timing decay coefficient; W S / W T / W ST The coupling weights obtained above; K S / KT / K ST The above refers to the dynamic coefficients.

[0060] In this embodiment, the coupling characteristic value of a single component at a single time node is obtained in the following way: The real-time S-value and T-value of each component in the target subarray are retrieved synchronously every 1 second, combined with fixed coupling weights and dynamic coefficients; For each time point (1 second / time), the coupling function described above is applied to each component for calculation. .

[0061] The subarray-level real-time coupling feature value sequence obtained based on the coupling feature value of each single component at a single time node includes: Aggregated hierarchically by component → group → subarray, all components within each group. .

[0062] Take the average value to obtain the string-level feature value; take the average value of all string-level feature values ​​to obtain the subarray-level feature value. Understandably, the feature value calculations of all components are strictly aligned with the time nodes (based on the unified timestamp of the power plant, with an error of ≤10ms) to avoid aggregation deviations caused by time misalignment; In this embodiment, generating a three-dimensional coupling matrix of subarray-timescale-timenode based on the subarray-level real-time coupling feature value sequence includes: Dimension definition and parameter settings First dimension (subarray dimension): If monitoring a single subarray (e.g., Z5), the dimension length = 1; if monitoring multiple subarrays (e.g., Z1-Z10), the dimension length = the number of subarrays. Dimension labels: Named according to subarray number (e.g., Z5, Z1-Z10). The second dimension (time scale dimension): It is fixedly divided into 3 scales: 10-second level (short-term fluctuations), 1-minute level (medium-term trends), and 10-minute level (long-term changes), with a dimension length of 3; Scale aggregation rules: 10-second level: The average of the feature values ​​of every 10 1-second nodes is taken to obtain a 10-second node value; 1-minute level: The feature values ​​of every 6 10-second nodes are averaged to obtain a 1-minute node value; 10-minute level: The average of the feature values ​​of every 10 1-minute nodes is taken to obtain a 10-minute node value; Dimension labels: 10s, 1min, 10min. The third dimension (time node dimension): Using the current 1-hour period as a cycle, the number of nodes is uniformly aligned according to the 10-second scale (360 nodes in 1 hour). 1-minute and 10-minute nodes are adapted through completion (e.g., 60 nodes for 1 minute, and the remaining 300 nodes fill the average value within that minute). Dimension label: timestamp. Matrix data filling: Data mapping: The sequence of subarray-level eigenvalues ​​is filled into the corresponding positions of the three-dimensional matrix according to the correspondence between time scale and time node, forming a three-dimensional coupling matrix of subarray-time scale-time node with a dimension of subarray number × 3 × 360. In this embodiment, the three-dimensional coupling matrix of subarray-time scale-time node is reduced in dimensionality using the Tucker decomposition algorithm to obtain the spatiotemporal nonlinear coupling matrix, including: Load Tucker decomposition parameters, including: Core tensor dimension: defaults to number of subarrays × 3 × 180; Factor matrices A (submatrix dimension factor matrix, dimension: number of submatrixes × number of submatrixes), B (time scale dimension factor matrix, dimension: 3×3), and C (time node dimension factor matrix, dimension: 360×180). Decomposition calculation: For a three-dimensional matrix X (dimension: I×J×K, I = number of subarrays, J = 3, K = 360), decompose it according to the Tucker decomposition formula, where G is the core tensor (dimension: I×3×180). Extract the factor matrix C of the time node dimension, project the three-dimensional matrix onto the two-dimensional space to obtain the two-dimensional matrix Y (dimension: I×360, submatrix - time node). Data normalization: Mapping the elements of the two-dimensional matrix to the [0,1] interval to ensure consistency with the input format of subsequent models; Information retention rate verification: Calculate the ratio of the Frobenius norm of the matrix before and after decomposition. If the ratio is <0.95 (i.e., information retention rate <95%), readjust the decomposition parameters (such as the core tensor dimension) and decompose again.

[0063] In this embodiment, the real-time prediction and deviation analysis of power generation based on the spatiotemporal nonlinear coupling matrix includes: Obtain the trained optical optimization matrix neural network; The spatiotemporal nonlinear coupling matrix is ​​input into the optical optimization matrix neural network to obtain the predicted power generation value for a future preset time period; Obtain historical power generation data for the same period; The deviation rate is calculated based on historical power generation data for the same period and the projected power generation for a future time period. The operating status is obtained based on the deviation rate; The correlation degree of deviation source tracing is obtained based on the deviation rate and the response intensity of each principal component.

[0064] In this embodiment, the optical optimization matrix neural network outputs a predicted power generation value for a preset future time period, including three time granularities: Short-term: Power generation forecast within 1 minute (1 node, corresponding to a 1-minute period); Medium term: Power generation forecast within 5 minutes (5 nodes, 1 minute / node); Long-term: Power generation forecast within 10 minutes (10 nodes, 1 minute / node).

[0065] In this embodiment, calculating the deviation rate based on historical power generation data for the same period and the predicted power generation for a preset future time period includes: Retrieve real-time power generation data for the same period (corresponding one-to-one with the predicted time period, such as the actual power generation of 10 nodes within 10 minutes). Calculate the deviation rate using the following formula: Where n is the number of prediction nodes (e.g., 10). Predict the power generation for the i-th node. This represents the actual power generation of the i-th node. The operational status is obtained based on the deviation rate, including: Normal state: Deviation rate <5%, no intervention required; Minor anomalies: 5% ≤ deviation rate < 10%, marked as to be observed, and the deviation changes will be continuously monitored for the next 3 cycles; Severe anomaly: Deviation rate ≥10%, triggering the anomaly diagnosis process.

[0066] In this embodiment, the correlation degree of deviation tracing can be obtained in the following way: Call the principal component response intensity (PC1 / PC2 / PC3) to calculate the correlation between each principal component and the power generation deviation (e.g., PC1 is related to sunlight, and a high correlation indicates that the deviation may originate from sunlight fluctuations). By combining real-time fluctuations of multiple indicators (such as a 15% decrease in light intensity within 10 minutes and a 3°C increase in component temperature), the main influencing indicators are identified by sorting them from high to low correlation.

[0067] In this embodiment, a deviation cause report can also be generated according to preset rules, for example: If the principal component with the highest correlation is PC1 (spatial illumination features), and the real-time illumination fluctuation is >10%, the reason for the judgment bias is the instantaneous fluctuation of illumination intensity; If the principal component with the highest correlation is PC2 (time-temperature characteristic), and the module temperature exceeds the historical average for the same period by 5℃, the reason for the deviation is that the module temperature is too high, resulting in power degradation.

[0068] In this embodiment, the application further includes: A multi-dimensional indicator anomaly diagnosis module is used to perform tiered investigation when the analysis results of the real-time power generation prediction and deviation analysis module exceed a preset safety threshold.

[0069] In this embodiment, when the analysis result of the real-time power generation prediction and deviation analysis module exceeds a preset safety threshold, a tiered investigation is performed, including: The safety threshold is determined as follows: If the power generation deviation rate is ≥10% (serious abnormality) and fails to recover after 2 consecutive cycles (2 minutes), a problem is identified and an alarm is triggered. Metric threshold trigger: Module level: Module current drop ≥50%, module temperature >45℃ (high temperature threshold), module power <70% of rated power; Inverter level: Inverter input / output power difference > 5%, power factor < 0.95 (qualified threshold), inverter temperature > 60℃ (safety threshold), abnormal operation status code display; Grid and environment level: Grid voltage > 10.5kV or < 9.5kV (rated voltage ± 5%), grid frequency > 50.2Hz or < 49.8Hz, wind speed > 15m / s (extreme wind speed threshold); Coupling matrix triggering: The cosine similarity between the two-dimensional coupling matrix and the standard matrix under the same operating conditions is <0.85 (consistency not met) and the mutual information value is <0.8 (low correlation with power generation).

[0070] In this embodiment, the hierarchical inspection includes component-level inspection, inverter-level inspection, and grid and environment-level inspection. Component-level inspection includes, for example: If the current of a single component is 0 and the voltage is normal (e.g., component Z5-C6-J10 I=0A, U=36.5V), combined with the neighborhood influence coefficient S<0.8, it is determined that the component circuit is open (e.g., the wiring terminals are loose). If the module temperature is >45℃ and the power is <80% of the rated power, and the light intensity is normal (I>800W / ㎡), it is determined to be "module hot spot failure" (such as caused by partial shading). If the component power is consistently less than 90% of the rated power and the power of neighboring components is normal (S≈1.0), combined with historical data (components have been in operation for more than 5 years), it is determined to be a "component power attenuation fault"; the investigation results are: output the component-level abnormality type (no abnormality / line open circuit / hot spot / power attenuation) and the specific component number. Inverter-level troubleshooting, for example: If the inverter input power is normal (deviation from the total power of the subarray components <2%) and the output power = 0, the status code indicates output overcurrent, which is determined to be a fault on the inverter output side (such as output switch tripping). If the power factor is <0.95 and the grid voltage is normal, combined with the inverter temperature being normal (<50℃), it is determined that the inverter control parameters are abnormal (such as a power factor adjustment module failure). If the inverter temperature is >60℃ and the fan speed is 0, and the input / output power drops by ≥20%, it is determined to be an inverter heat dissipation failure (such as fan damage). Troubleshooting results: Output inverter-level anomaly type (no anomaly / output side fault / control parameter anomaly / heat dissipation fault) and specific inverter number. Power grid and environmental level inspections, for example: If the grid voltage is >10.5kV or <9.5kV, and the power generation of all subarrays decreases by ≥15%, it is determined to be an abnormal fluctuation in grid voltage. If the total harmonic distortion (THD) of the power grid is greater than 5% and the power factor fluctuation of the inverter is greater than 0.05, it is determined that the power grid harmonics exceed the standard. If the wind speed is greater than 15 m / s and displacement monitoring data is found on the component support, it is determined that the component displacement is caused by extreme wind speed. Investigation results: Output power grid and environmental anomaly types (no anomaly / voltage fluctuation / harmonic exceedance / extreme wind speed). In this embodiment, fault location and report generation can also be performed, and the specific scheme is as follows: Based on the combined results of the three-tier investigation, the final anomaly type was determined according to the priority of component level > inverter level > grid and environment level (component-level failures have a more direct impact); If the component-level investigation reveals a line break, and there are no abnormalities at the inverter level or the grid level, the final location is a line break in the Z5-C6-J10 component. The report may include the following: Basic information about the anomaly: anomaly trigger time, involved subarray / device number, and anomaly duration; Fault details: anomaly type, specific location, and scope of impact (e.g., affecting 5% of the power generation of the Z5 subarray); Urgency level: High (e.g., immediate shutdown required due to line breakage), Medium (e.g., troubleshooting required within 24 hours due to abnormal parameters), Low (e.g., environmental factors require observation). Preliminary troubleshooting suggestions: Stop the machine and check the wiring terminals of the Z5-C6-J10 component. Prepare insulated gloves and a screwdriver.

[0071] In this embodiment, the above-mentioned optical matrix neural network model can adopt a basic architecture of a fully connected neural network (MLP) (corresponding network: multilayer perceptron) + an enhancement architecture, wherein the enhancement architecture is an attention mechanism + MLP (corresponding network: fully connected network with attention, the core attention module adopts a compression-excitation (SE) structure).

[0072] In this embodiment, each of the above models needs to undergo specific pre-training. The specific training data acquisition method is as follows: Acquire training data covering component-level, inverter-level, grid-level, environmental-level, and power generation data, consistent with the dimensions collected in the real-time phase, to ensure data correlation; Module level: Temperature, current, voltage, and light intensity data of all modules over the past 1-3 years, with a time granularity of ≤1 minute (to ensure the capture of short-term fluctuations). Inverter level: synchronous inverter input / output power, power factor, temperature, and operating status code, with a time granularity of ≤1 minute; Grid level: Grid-connected voltage, frequency, harmonic content, time granularity ≤ 5 minutes; Environmental grade: ambient temperature, humidity, wind speed, wind direction, time granularity ≤ 5 minutes; Power generation: Total / subarray / module-level power generation, time granularity ≤ 1 minute; Operating conditions coverage: It must include 27 core operating conditions (based on a combination of 3 levels of light intensity × 3 levels of component temperature × 3 levels of shading rate), with ≥3000 sets of data for each operating condition (to ensure statistical significance), and ≥500 sets of data for special operating conditions (such as heavy rain, heavy snow, and strong shading).

[0073] Understandably, the above data also needs to be preprocessed, such as basic operations like missing value imputation, noise reduction, and standardization.

[0074] In this embodiment, the response scale rule base can be constructed in the following way: Component-level core feature calculation: For each historical data point, the neighborhood influence coefficient, temporal decay coefficient, and three types of cross-features (spatial illumination, temporal temperature, and spatiotemporal power) are calculated using the same logic as in the real-time stage. Statistically analyze the distribution patterns of each feature under different operating conditions (e.g., the neighborhood influence coefficient is mostly <0.8 under occlusion conditions) to determine the effective range of feature values; Multiscale factor calculation: The spatial scale factor (the ratio of the coefficient of variation at the component level to that at the subarray level) is calculated for each working condition based on a three-level spatial scale of component-string-subarray. Calculate the time scale factor (ratio of 10-second to 10-minute fluctuations) for each operating condition using three time scales: 10 seconds, 1 minute, and 10 minutes. Calculate the spatiotemporal crossover scale factor (spatial scale factor × temporal scale factor × spatiotemporal-power crossover feature). Principal component response intensity calculation: KPCA dimensionality reduction was performed on the cross features of all historical data (RBF kernel function was selected, and gamma value was optimized through cross-validation) to extract three principal components (corresponding to spatial illumination, temporal temperature, and spatiotemporal power features, respectively). Calculate the average response intensity of the three principal components under each working condition label to form a working condition-principal component response intensity mapping table (e.g., for working conditions I high-T medium-C low, principal component 1 = 0.78, principal component 2 = 0.42, and principal component 3 = 0.25). Principal component response intensity grading: Three threshold levels—high (>70th quantile), medium (30%-70th quantile), and low (<30th quantile)—are determined based on distribution quantiles to form a grading standard. Rule base coefficient optimization training: Rule combination definition: Based on 3 principal component response intensity levels (high / medium / low) × 3 spatial scale factor levels (high / medium / low) × 3 temporal scale factor levels (high / medium / low), 81 rule combination scenarios are formed; Iterative optimization of weight formula coefficients: A weight formula framework is preset for each combination scenario, consistent with the actual use stage; With the goal of minimizing the predicted deviation of power generation after weight calculation in this scenario, the particle swarm optimization algorithm is used to optimize the coefficients A / B / C / D / E / F (the constraint coefficients are in the range of 0.05-0.5 to avoid a single coefficient dominating). Introduce photovoltaic characteristic constraints: such as forcing A≤0.3 under shading conditions (weakening the influence of space factor) and forcing D≥0.15 under high temperature conditions (enhancing the influence of time-temperature characteristics).

[0075] The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of this application has the following advantages: (1) In the calculation of core features, the neighborhood influence coefficient is dynamically adjusted according to the component layout (plane / sloping surface) range (3×3 component matrix for plane, 2 components at the top and bottom for sloping surface), and the time-series decay coefficient is updated in real time according to the power trend within a 10-minute time window to avoid the problem that fixed parameters cannot be adapted to different installation scenarios or time trends. When calculating the coupling weights, the exclusive coefficients are matched from the response scale rule base based on the real-time classification of principal component response intensity (spatial illumination / temporal temperature / spatial-temporal power) and multi-scale factors (spatial / temporal / spatial-temporal intersection), instead of using a fixed weight formula; In the dynamic coefficient generation stage, the coefficients output by the GBT model are adapted and adjusted according to the current operating conditions (such as shading and high temperature) (limiting the spatial coefficient when shading and limiting the spatiotemporal cross coefficient when high temperature) to ensure that the coefficients conform to the physical operation law of photovoltaic equipment. This dynamic adaptation design enables the solution to accurately cope with complex operating conditions such as fluctuations in sunlight, temperature changes, and partial shading in photovoltaic power plants. When there are frequent fluctuations in sunlight during cloudy weather, the time-series attenuation coefficient and dynamic coefficient can be adjusted in real time to follow the power trend, avoiding the characteristic calculation deviation caused by fixed parameters. In scenarios where roof edge components are susceptible to shading, the dynamic range adjustment of the neighborhood influence coefficient can accurately capture the impact of shading on component power, providing reliable feature data for subsequent anomaly diagnosis and efficiency optimization. Ultimately, this achieves the stability of analysis results under different operating conditions, avoids a sudden drop in accuracy or analysis failure due to changes in operating conditions, and ensures the reliability of operation and maintenance decisions.

[0076] (2) By fusing component-level (temperature, current), environmental-level (light, temperature and humidity), and subarray-level (average power) data through cross-feature (spatial illumination, temporal temperature and temperature, spatiotemporal power), cross-dimensional correlations (such as the coupling effect of light intensity and component spatial location, temperature change and time decay trend) are quantified. In the linkage analysis, the source tracing of power generation prediction deviations combines multi-dimensional real-time indicators and principal component response intensity correlation ranking. By using the correlation degree to locate the main influencing factors that cause deviations (such as light fluctuations and high module temperature), rather than relying solely on a single power indicator. The anomaly diagnosis adopts a hierarchical troubleshooting logic of component level → inverter level → grid and environment level, and integrates multi-dimensional fault characteristics (such as sudden drop in component current + no abnormality in inverter + normal grid voltage → locating component line fault), avoiding misjudgment or omission caused by single-dimensional judgment. In the analysis of power generation prediction deviation, the core cause of the deviation can be identified by ranking the correlation (such as the deviation being dominated by light fluctuations). Maintenance personnel do not need to blindly investigate and can take targeted measures (such as cleaning dust from the surface of the components to improve the utilization rate of sunlight). During anomaly diagnosis, layered investigation combined with multi-dimensional features can accurately locate faulty equipment (such as a circuit problem in a specific component), rather than just locating it to the subarray or system level, reducing the scope and workload of maintenance personnel. The overall analysis results are no longer single numerical values ​​or vague conclusions, but interpretable reports with influencing factors and related logic, helping operations and maintenance personnel understand the source of the analysis results and increasing their trust in decision-making.

[0077] (3) This application can support long-term optimization by constructing a data-driven mechanism of operation-feedback-iteration. Specifically, during real-time operation, data such as the results of abnormal handling (e.g., the recovery of power generation after fault repair) and the effect of efficiency optimization (e.g., the temperature and power changes after starting the cooling fan) will be automatically transmitted back to the historical database. This feedback data can be used to update core resources during the training phase (such as coefficients of the response-scale rule base, weights of the optical optimization matrix neural network, and historical best index database), while supplementing training samples for new operating conditions (such as extreme high temperatures and continuous rain). Regularly (e.g., quarterly) iterate and optimize the model and rule base based on feedback data to ensure that the solution continues to adapt to long-term factors such as equipment aging and environmental changes as the power plant operates for longer periods. In the early stages of power plant operation, the solution can provide basic analytical capabilities based on initial training data; as the operation time increases, the feedback data becomes richer, the model and rule base are continuously optimized, and the analytical accuracy and adaptability gradually improve. In response to the power degradation of components caused by equipment aging, the historical best index database can dynamically update the optimal operating parameters of aging components, and efficiency optimization suggestions will also be adjusted accordingly (such as gradually transitioning from starting the cooling fan to component replacement suggestions). This iterative mechanism ensures that the solution always aligns with the actual operating status of the power plant, continuously providing accurate analysis and optimization support for operation and maintenance, helping the power plant maintain high power generation efficiency throughout its entire life cycle, and achieving long-term economic benefits.

[0078] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A distributed photovoltaic power generation capacity real-time monitoring and multi-dimensional operation index linkage analysis system, characterized in that, The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system comprises: A current data acquisition module, which is configured to acquire distributed photovoltaic power generation real-time monitoring data; A core feature calculation module, which is configured to calculate component-level core features, multi-scale factors and principal component response intensities according to the distributed photovoltaic power generation real-time monitoring data; A coupling weight and dynamic coefficient calculation module, which is configured to generate coupling weights and dynamic coefficients according to the component-level core features, the multi-scale factors and the principal component response intensities; A spatio-temporal nonlinear coupling matrix generation module, which is configured to generate a spatio-temporal nonlinear coupling matrix according to the coupling weights, the dynamic coefficients, the component-level core features, the multi-scale factors and the principal component response intensities; A power generation real-time prediction and deviation analysis module, which is configured to perform power generation real-time prediction and deviation analysis according to the spatio-temporal nonlinear coupling matrix. 2.The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of claim 1, wherein, The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system further comprises: A multi-dimensional index abnormality diagnosis module, which is configured to perform hierarchical troubleshooting when the analysis result of the power generation real-time prediction and deviation analysis module exceeds a preset safety threshold. 3.The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of claim 2, wherein, The power generation real-time monitoring data comprises component-level data, inverter-level data, power grid-level data, environmental-level data and power generation data. 4.The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of claim 3, wherein, The component-level core features are calculated by the following steps: A neighborhood influence coefficient is calculated according to the power generation real-time monitoring data; A time-series attenuation coefficient is calculated according to the power generation real-time monitoring data; A spatial light cross feature is calculated according to the power generation real-time monitoring data and the neighborhood influence coefficient; A time-temperature cross feature is calculated according to the power generation real-time monitoring data and the time-series attenuation coefficient; A spatio-temporal power cross feature is calculated according to the power generation real-time monitoring data. 5.The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of claim 4, wherein, The multi-scale factors are calculated by the following steps: Component-level, string-level and sub-array-level variation coefficients are calculated with the neighborhood influence coefficient as the calculation object; A spatial scale factor is obtained according to the component-level variation coefficient and the sub-array-level variation coefficient; A time scale factor is calculated according to the time-series attenuation coefficient; A spatio-temporal cross-scale factor is calculated according to the spatial scale factor, the time scale factor and the spatio-temporal power cross feature. 6.The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of claim 5, wherein, The principal component response intensities are obtained by the following method: A trained KPCA dimension reduction model is obtained; The spatial light cross feature, the time-temperature cross feature and the spatio-temporal power cross feature are dimensionally reduced by the trained KPCA dimension reduction model, three principal components are extracted, and the principal component response intensities corresponding to the three principal components are obtained. 7.The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of claim 6, wherein, The spatio-temporal nonlinear coupling matrix is generated according to the coupling weights, the dynamic coefficients, the component-level core features, the multi-scale factors and the principal component response intensities, which comprises: A response scale rule library is obtained; The principal component response intensities corresponding to each principal component are classified by intensity, thereby obtaining the corresponding level of each principal component; The spatial scale factor, the time scale factor, and the space-time cross scale factor are graded to obtain a spatial scale factor level, a time scale factor level, and a space-time cross scale factor level; The weight formula coefficients are obtained from the response scale rule library according to the corresponding level of each principal component, the spatial scale factor level, the time scale factor level, and the space-time cross scale factor level; The spatial weight initial value, the time weight initial value, and the space-time cross weight initial value are obtained according to the obtained weight formula coefficients; The spatial weight initial value, the time weight initial value, and the space-time cross weight initial value are calibrated; A trained dynamic coefficient generation model is obtained; The dynamic coefficients including the spatial coefficient, the time coefficient, and the space-time cross coefficient are obtained by inputting the principal component response intensity, the spatial scale factor, the time scale factor, the space-time cross scale factor, the calibrated spatial weight initial value, the calibrated time weight initial value, and the calibrated space-time cross weight initial value into the trained dynamic coefficient generation model. 8.The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of claim 7, wherein, The space-time nonlinear coupling matrix is generated according to the coupling weight, the dynamic coefficient, the component-level core feature, the multi-scale factor, and the principal component response intensity, which includes: A preset coupling function is obtained; The coupling weight, the dynamic coefficient, the component-level core feature, the multi-scale factor, and the principal component response intensity are substituted into the preset coupling function to obtain the coupling feature value of a single component and a single time node; The sub-array-level real-time coupling feature value sequence is obtained according to the coupling feature value of each single component and single time node; The sub-array-time scale-time node three-dimensional dimension coupling matrix is generated according to the sub-array-level real-time coupling feature value sequence; The space-time nonlinear coupling matrix is obtained by performing dimension reduction processing on the sub-array-time scale-time node three-dimensional dimension coupling matrix through a Tucker decomposition algorithm. 9.The distributed photovoltaic power generation real-time monitoring and multi-dimensional operation index linkage analysis system of claim 8, wherein, The real-time power generation prediction and deviation analysis are performed according to the space-time nonlinear coupling matrix, which includes: A trained light optimization matrix neural network is obtained; The space-time nonlinear coupling matrix is input into the light optimization matrix neural network to obtain the power generation prediction value of a future preset time period; Contemporary historical power generation data are obtained; The deviation rate is calculated according to the contemporary historical power generation data and the power generation prediction value of the future preset time period; The running state is obtained according to the deviation rate; The deviation traceability correlation degree is obtained according to the deviation rate and the principal component response intensity.

10. A method for real-time monitoring of power generation of distributed photovoltaics and multi-dimensional operation index linkage analysis, characterized in that, The real-time monitoring and multi-dimensional running index linkage analysis method of the distributed photovoltaic power generation includes: Real-time monitoring data of the distributed photovoltaic power generation are obtained; The component-level core feature, the multi-scale factor, and the principal component response intensity are calculated according to the real-time monitoring data of the distributed photovoltaic power generation; The coupling weight and the dynamic coefficient are generated according to the component-level core feature, the multi-scale factor, and the principal component response intensity; The space-time nonlinear coupling matrix is generated according to the coupling weight, the dynamic coefficient, the component-level core feature, the multi-scale factor, and the principal component response intensity; The real-time power generation prediction and deviation analysis are performed according to the space-time nonlinear coupling matrix.

Images