An all-scene photovoltaic power intelligent simulation and risk management and control system and method

Abstract

The present application relates to the technical field of new energy power system intelligence, and particularly relates to a full-scene photovoltaic power intelligent simulation and risk management and control system and method. The system comprises a basic simulation module, a data fusion and feature extraction module, a working condition recognition module with multi-level confidence layering judgment, a two-level adaptive parameter adjustment module, a chain disaster coupling simulation module and a proactive hierarchical risk early warning module. The present application uses a multi-level threshold anti-shake logic to complete multi-type working condition layering recognition and archive unknown working condition samples; compensates for photovoltaic output drift through a two-level correction mechanism of working condition gear coarse adjustment combined with amplitude limiting PID fine adjustment; deduces chain disaster power loss using Monte Carlo sampling, dynamic disaster weighting factor and time decay function; and outputs standardized hierarchical early warning relying on a risk matrix. The present application constructs a photovoltaic power simulation full-process intelligent closed-loop management and control system, improves simulation adaptive accuracy, composite working condition and extreme chain disaster simulation capability and preposition risk early warning landing.

Description

Technical Field

[0001] This invention relates to the field of intelligent technology for new energy power systems, and in particular to a full-scenario intelligent simulation and risk management system and method for photovoltaic power. Specifically, it relates to a full-scenario intelligent simulation and risk management system and method for photovoltaic power that integrates multi-source data fusion, artificial intelligence operating condition recognition, dynamic parameter self-adaptation, chain disaster coupling simulation, and proactive early warning and control. Background Technology

[0002] Photovoltaic power generation is an important component of the clean energy system. However, photovoltaic output is inherently intermittent and fluctuating, and large-scale grid connection can easily lead to power imbalances, voltage fluctuations, and other safety and stability issues in the power grid. High-precision photovoltaic power simulation and proactive risk assessment throughout the entire process are core technological supports for improving the predictability of photovoltaic output, realizing intelligent operation and maintenance of power plants, and building a proactive safety defense system for the power system. Currently, the industry has developed various photovoltaic power simulation, online monitoring, and fault diagnosis technology solutions. However, existing technologies still have unavoidable technical shortcomings in areas such as full-scenario adaptive simulation, refined identification of multiple operating conditions, coupled simulation of extreme disasters, and linkage between simulation and risk management.

[0003] Current photovoltaic (PV) operating condition identification technologies rely on a single judgment method and lack a probabilistic confidence quantification evaluation system. This makes it impossible to distinguish between multiple overlapping and complex operating conditions and unknown abnormal operating conditions outside the sample domain. For example, the PV anomaly identification algorithms disclosed in existing patents CN119397927A or CN121937094A only output binary judgment results of normal or faulty conditions or single operating condition labels, without outputting probabilistic confidence scores that characterize the reliability of the identification. This makes it impossible to quantify the reliability of the operating condition identification results. Such technical solutions can only identify single, independent operating conditions and cannot distinguish between multiple complex operating conditions where dust accumulation, partial shading, and cloud disturbance coexist. Furthermore, they lack dedicated identification and archiving processes for entirely new and unknown abnormal operating conditions not covered by the training samples, simply categorizing them as "other." The accuracy of the operating condition identification output results is insufficient, failing to provide a reliable input data source for simulation parameter correction and disaster risk projection, and also failing to accumulate new abnormal samples to support system iterative optimization.

[0004] Existing photovoltaic simulation model parameter adaptive adjustment methods are fixed and lack a closed-loop fine correction architecture that combines layered coarse and fine adjustments. For example, the photovoltaic parameter adjustment methods provided by existing patents CN121412901A or CN121744870A adopt one-time overall correction or fixed linear rule coarse adjustment, which can only achieve large-span parameter changes and is difficult to continuously track the output performance drift caused by long-term gradual factors such as photovoltaic module aging, surface dust accumulation, and partial shading. The simulation output error continues to accumulate with the operation of the equipment. Existing technologies have not constructed a two-level adaptive parameter tuning architecture that combines coarse adjustment based on operating condition mapping and fine-tuning using error closed-loop PID. They lack a smooth buffer mechanism for parameter switching and upper and lower limit constraints for parameters. During the parameter adjustment process, problems such as abrupt changes in the simulation power curve and model parameters exceeding limits and violating photovoltaic physical characteristics are prone to occur, making it impossible to maintain stable simulation accuracy in the long term.

[0005] Current extreme weather simulations only calculate single disaster events independently, without establishing a nonlinear coupling calculation model for time-series chain disasters. For example, existing patent CN122178281A achieves meteorological simulation by discretely injecting extreme events, treating extreme weather events such as high temperature, dust storms, rainfall, and strong winds as unrelated static and independent scenarios for separate calculations. It simply superimposes multiple types of disaster parameters without reproducing the nonlinear synergistic attenuation chain coupling effect caused by the time-series superposition of multiple disasters. Taking a dust storm and rainfall chain disaster as an example, rainfall will solidify dust on the surface of photovoltaic panels, forming mud stains, which will continuously aggravate the attenuation of irradiance, and this secondary attenuation effect will gradually dissipate over time. Existing technologies do not configure disaster weighting factors that dynamically change with the number of superimposed disasters, nor piecewise time attenuation functions that monotonically decrease with the simulation duration. They cannot fully quantify the power loss of the entire chain disaster process, resulting in large deviations in the risk assessment results of extreme scenarios, making it difficult to support the preparation of refined disaster prevention and maintenance plans.

[0006] Currently, photovoltaic simulation modules and risk warning modules operate independently, with fragmented data interaction links, failing to form an integrated intelligent closed loop encompassing operational condition diagnosis, power prediction, and tiered decision-making. For example, existing patent CN119204639A relies on fixed static thresholds to classify risk levels, referencing only historical statistical data for risk assessment without establishing a direct mapping relationship with quantitative indicators such as power loss rate and estimated power outage duration output by dynamic simulation. Such technical solutions only vaguely identify risk levels in their warning information, failing to correlate affected equipment or quantify the scale of power loss, and lack standardized tiered operation and maintenance guidelines. Simulation calculation, risk assessment, and operation and maintenance management are isolated information silos, lacking a complete business closed loop of "operational condition diagnosis - disaster power prediction - tiered risk decision-making," resulting in insufficient intelligent linkage support capabilities.

[0007] In summary, existing photovoltaic power simulation and related risk management technologies suffer from multiple technical shortcomings, including the inability to quantify the reliability of operating condition identification, a lack of ability to identify complex operating conditions and unknown anomalies, a single adaptive parameter adjustment mechanism, a lack of chain-like extreme disaster coupled simulation modeling, and a disconnect between simulation and risk early warning systems. There is an urgent need for an integrated intelligent photovoltaic power simulation and risk management system. This system should possess comprehensive capabilities including full-condition confidence-level hierarchical identification, two-level hierarchical adaptive parameter correction, dynamic coupled extrapolation of chain-like disasters, and automatic mapping of simulation results to hierarchical management strategies. This would improve the intelligent operation and maintenance level of photovoltaic power plants and strengthen the power system's new energy security defense capabilities. Summary of the Invention

[0008] This invention aims to overcome the aforementioned deficiencies of existing technologies and provide a full-scenario intelligent photovoltaic power simulation and risk management system and method. This system can not only accurately simulate the output characteristics of photovoltaic power plants under various complex and time-varying operating conditions, but also achieve intelligent diagnosis of operating status, chain-coupled simulation of extreme disasters, and generate accurate and operable proactive risk warnings, forming an intelligent operation and maintenance process from data perception to decision-making closed loop.

[0009] According to a first aspect of the present invention, a full-scenario photovoltaic power intelligent simulation and risk management system is provided, comprising: a basic simulation module for performing power simulation based on a photovoltaic physical mechanism model and simulation parameters, and outputting simulated power data; a data fusion and feature extraction module configured with an offline simulation working mode and an online operation and maintenance working mode; in the offline simulation working mode, collecting simulated power data, meteorological data, and power plant operating status data; in the online operation and maintenance working mode, collecting simulated power data, real-time on-site power data, meteorological data, and power plant operating status data; in both the offline simulation working mode and the online operation and maintenance working mode, extracting features through a sliding window and constructing a multi-dimensional time-series feature vector; and a working condition identification module with multi-level confidence hierarchical judgment, used to input the multi-dimensional time-series feature vector into a multi-label probability classification model, outputting confidence levels for various working conditions, and combining multi-level threshold anti-shake logic with configured multi-level confidence thresholds, differentiated continuous judgment periods, and working condition mutual exclusion priorities to determine the working condition, the working condition types including single conventional working conditions, etc. The system includes simulations of conventional composite operating conditions, single extreme operating conditions, chain-like composite disaster operating conditions, conventional-extreme cross-composite operating conditions, and unknown operating conditions. A two-level adaptive parameter tuning module is used to adaptively adjust simulation parameters based on operating condition identification results. This includes coarse-tuning of power levels using an operating condition-parameter mapping strategy, and fine-tuning of PID with parameter-limited constraints based on the deviation between simulated power data and real-time on-site power data. A chain-like disaster coupling simulation module receives the judgment results of single extreme operating conditions or chain-like composite disaster operating conditions, retrieves historical measured extreme weather data matching the currently received operating conditions, extracts extreme time-series parameters using Monte Carlo random sampling, and performs coupled calculations using disaster weighting factors and time decay functions. This data is then input into the basic simulation module to deduce the power impact of chain-like disasters. The extreme time-series parameters are meteorological dynamic subsets from the simulated power data. An active graded risk warning module retrieves simulation results of the power impact of chain-like disasters, matches them with a multi-dimensional risk matrix, and automatically generates standardized graded warning information.

[0010] Furthermore, in the data fusion and feature extraction module, the step size and window length of the sliding window are adaptively configured according to the installed capacity of the power plant and the data acquisition frequency. The multi-dimensional time-series feature vector includes: power time-domain features taken from simulated power data or real-time power data on site, which include power mean, standard deviation, peak factor, skewness coefficient, and slope mean; meteorological features taken from meteorological data, which include the mean, maximum, minimum, rate of change, and trend inflection point of irradiance, temperature, wind speed, and dust concentration within the same sliding window; and status labels taken from power plant operating status data, which are discrete status markers for equipment fault alarms, inverter start / stop, and string disconnection.

[0011] Furthermore, the multi-level threshold anti-shake logic is configured with four types of confidence thresholds: a regular high-confidence threshold, an extreme high-confidence threshold, a general medium-confidence threshold, and an unknown low-confidence threshold. The numerical relationship of these four types of confidence thresholds is as follows: the regular high-confidence threshold is greater than the extreme high-confidence threshold, the extreme high-confidence threshold is greater than the general medium-confidence threshold, and the general medium-confidence threshold is greater than the unknown low-confidence threshold. The multi-level threshold anti-shake logic is configured with six types of continuous judgment cycles specific to different operating conditions: a single regular judgment cycle, a regular composite judgment cycle, a single extreme judgment cycle, a chain-like composite disaster judgment cycle, a regular-extreme cross-composite judgment cycle, and an unknown operating condition judgment cycle. The duration relationship of the first five types of continuous judgment cycles is as follows: a single regular judgment cycle... The fixed-cycle period is longer than the conventional composite judgment cycle, the conventional composite judgment cycle is longer than the conventional-extreme cross-composite judgment cycle, the conventional-extreme cross-composite judgment cycle is longer than the single extreme judgment cycle, the single extreme judgment cycle is equal in length to the chain-type composite disaster judgment cycle, and the unknown working condition judgment cycle adopts an independent duration configuration and is not included in the duration ranking comparison of the first five types of working conditions; the multi-level threshold anti-shake logic presets the working condition judgment priority order, from high to low as chain-type composite disaster, single extreme, conventional-extreme cross-composite, conventional composite, and single conventional; if the high-priority valid working condition judgment condition is met, the judgment process of all low-priority valid working conditions is blocked; the unknown working condition judgment process is only executed when all valid working condition judgment conditions are not met.

[0012] Furthermore, the multi-level threshold debouncing logic is configured as follows: If, without triggering a higher-priority valid operating condition, the confidence level of any regular operating condition label remains consistently higher than the regular high-confidence threshold within a single regular judgment period, it is determined to be a single regular operating condition; if no regular operating condition meets the regular high-confidence single operating condition condition, and the confidence levels of two or more regular operating condition labels remain consistently higher than the general medium-confidence threshold within a regular composite judgment period, it is determined to be a regular composite operating condition; if the confidence level of any single extreme operating condition label remains consistently higher than the extreme high-confidence threshold within a single extreme judgment period, it is determined to be a single extreme operating condition, and all subsequent judgments of regular operating conditions are terminated; if no single extreme operating condition meets the extreme high-confidence single-operation condition condition... Under certain conditions, if the confidence levels of two or more single extreme condition labels remain consistently above the general medium confidence threshold during the chain-compound disaster determination period, it is determined to be a chain-compound disaster condition, and all low-priority valid condition determinations are terminated. If, under the premise that no corresponding high-confidence single condition triggers either the normal or extreme condition categories, the confidence levels of at least one normal condition label and one extreme condition label remain consistently above the general medium confidence threshold during the normal-extreme cross-compound determination period, it is determined to be a normal-extreme cross-compound condition. If the confidence levels of all known condition labels remain below the unknown low confidence threshold throughout the unknown condition determination period, it is determined to be an unknown condition, and the corresponding original data and multi-dimensional feature vectors are automatically archived to the anomaly sample database.

[0013] Furthermore, the pre-stored working condition-parameter mapping strategy includes irradiance attenuation and photoelectric conversion efficiency correction levels corresponding to dust accumulation, shading, and aging scenarios. A smooth transition buffer is set during level switching to avoid abrupt changes in simulation parameters. The deviation judgment logic of the PID fine-tuning correction is as follows: when the relative error between the simulated power and the real-time power data on site exceeds the set threshold for two consecutive acquisition cycles, the PID controller is activated to perform small-step parameter correction, and the parameter amplitude constraint is a preset fixed value range.

[0014] Furthermore, the disaster weighting factor dynamically increases with the number of disasters superimposed, and the time decay function is a piecewise function that monotonically decreases with the simulation duration, used to characterize the gradual dissipation of the secondary impact of disasters over time.

[0015] Furthermore, the multidimensional risk matrix uses power loss rate and expected power outage duration as its two core dimensions, matching corresponding early warning levels and graded operation and maintenance guidelines.

[0016] Furthermore, it also includes a self-learning iteration module, which is used to periodically read archived data in the abnormal sample library. After confirming or supplementing the working condition labels of the archived data, it incrementally trains the multi-label probabilistic classification model and synchronously updates the gear parameters of the working condition-parameter mapping strategy.

[0017] Furthermore, it also includes a visualization interaction module, which provides a working condition display panel, an early warning history log window, and a multi-mode switching button, and is configured with interactive operation functions such as view freeze, local waveform magnification, and batch export of test cases.

[0018] According to a second aspect of the present invention, a method for intelligent simulation and risk management of photovoltaic power across all scenarios is provided, comprising: performing power simulation based on a photovoltaic physical mechanism model and simulation parameters through a basic simulation module, and outputting simulated power data; configuring an offline simulation working mode and an online operation and maintenance working mode through a data fusion and feature extraction module; collecting simulated power data, meteorological data, and power plant operating status data in the offline simulation working mode, and collecting simulated power data, real-time on-site power data, meteorological data, and power plant operating status data in the online operation and maintenance working mode; extracting features and constructing a multi-dimensional time-series feature vector through a sliding window in both the offline simulation working mode and the online operation and maintenance working mode; inputting the multi-dimensional time-series feature vector into a multi-label probability classification model through a working condition identification module with multi-level confidence hierarchical judgment, outputting the confidence level of various working conditions, and determining the working condition by combining multi-level threshold anti-shake logic with configured multi-level confidence thresholds, differentiated continuous judgment periods, and working condition mutual exclusion priorities, wherein the working condition type includes a single conventional working condition. The simulation is designed to simulate various operating conditions, including conventional composite operating conditions, single extreme operating conditions, chain-like composite disaster operating conditions, conventional-extreme cross-composite operating conditions, and unknown operating conditions. A two-level adaptive parameter adjustment module adjusts simulation parameters based on the operating condition identification results. This includes coarse adjustment of power levels using an operating condition-parameter mapping strategy, and fine-tuning of PID control with parameter limiting constraints based on the deviation between simulated power data and real-time on-site power data. A chain-like disaster coupling simulation module receives the judgment results of single extreme operating conditions or chain-like composite disaster operating conditions, retrieves historical extreme weather data matching the currently received operating conditions, extracts extreme time-series parameters using Monte Carlo random sampling, and performs coupled calculations using disaster weighting factors and time decay functions. These parameters are then input into the basic simulation module to deduce the power impact of chain-like disasters. The extreme time-series parameters are meteorological dynamic subsets from the simulated power data. An active graded risk warning module retrieves the simulation results of the power impact of chain-like disasters, matches them with a multi-dimensional risk matrix, and automatically generates standardized graded warning information.

[0019] Compared with existing photovoltaic power simulation, operating condition identification and risk management technologies, this invention forms a complete closed-loop technology system through a multi-level confidence-layered anti-shake operating condition identification architecture, a two-level layered adaptive parameter correction mechanism, a chain-like disaster coupling simulation calculation system, a two-dimensional multi-dimensional risk matrix, an abnormal sample incremental self-learning module, and an integrated visualization interaction module. Each module produces the following objective technical effects.

[0020] 1. This invention sets differentiated multi-level confidence thresholds, time-varying judgment periods, and mutually exclusive priorities for operating conditions, along with standardized multi-category operating condition judgment rules. It can output the probability confidence scores corresponding to each operating condition label, distinguishing between six types of operating conditions: single normal, multiple compound, single extreme, chain compound disaster, normal-extreme cross-compound, and unknown. Simultaneously, it automatically archives and stores the original data and feature vectors for unknown operating conditions. This structure overcomes the shortcomings of existing technologies that only output binary judgment results or single operating condition labels, cannot identify compound operating conditions, and lack a mechanism for collecting unknown abnormal samples. Through differentiated judgment periods of varying lengths and multi-level threshold anti-shake mechanisms, it filters out misjudgments caused by instantaneous lighting disturbances, improving the stability of operating condition recognition results. The collected unknown operating condition samples can serve as a data source for model iteration training, continuously expanding the coverage of operating condition recognition.

[0021] 2. This invention constructs a two-stage linkage parameter tuning architecture: a coarse adjustment based on the operating condition-parameter mapping and a fine-tuning based on the error closed-loop PID control. The gear switching configuration includes a smooth transition buffer, and all adjustable simulation parameters are set with fixed-range amplitude limits. This structure overcomes the shortcomings of existing technologies that employ single-cycle overall correction, single linear coarse adjustment, and lack parameter boundary constraints. The coarse adjustment stage can quickly match the attenuation correction gear based on the identified operating conditions, shortening the model adaptation response time. The PID fine-tuning stage initiates small-step correction based on multi-cycle continuous out-of-tolerance conditions, continuously compensating for output performance drift caused by component dust accumulation, aging, and shading. The smooth buffer and parameter limiting mechanism avoids problems such as instantaneous power jumps in simulation and parameter over-limit violations of photovoltaic physical constraints, ensuring stable control of the deviation range between simulation output and actual output during long-term operation.

[0022] 3. This invention employs a Monte Carlo random sampling algorithm to extract extreme time-series parameters. It combines this with a disaster weighting factor that dynamically increases with the number of superimposed disasters and a piecewise time decay function that monotonically decreases with the simulation duration to complete chain-like disaster coupling calculations. This comprehensively simulates the time-series changes in photovoltaic power under multiple superimposed disasters and outputs quantitative indicators such as power loss rate and expected power outage duration. This structure overcomes the shortcomings of existing technologies that only independently simulate single extreme weather events, simply superimpose disaster parameters, and fail to characterize the nonlinear secondary coupling effects of disasters. The coupling calculation process closely follows the physical laws of component attenuation under the cascading effects of sandstorms, rainfall, and high temperatures, improving the objectivity of power loss quantification results in extreme disaster scenarios and providing quantitative data support for disaster prevention and maintenance plan development.

[0023] 4. This invention constructs a multi-dimensional risk matrix with power loss rate and expected power outage duration as its two core dimensions. It establishes a fixed mapping relationship between chain-like disaster simulation quantitative indicators and risk levels and tiered operation and maintenance guidelines, automatically generating tiered early warning information that includes risk level, affected operating conditions, quantified power loss values, equipment safety risks, and standardized operation and maintenance guidelines. This structure overcomes the shortcomings of existing simulation and risk assessment modules, such as data fragmentation, reliance on static thresholds for risk judgment, lack of quantitative indicators to support early warning information, and a lack of targeted response plans. It establishes a data interaction link between operating condition identification, power simulation, risk assessment, and operation and maintenance decision-making, forming a complete business closed loop of "operating condition diagnosis - disaster power prediction - tiered risk decision-making."

[0024] 5. This invention incorporates a self-learning iterative module that periodically reads archived data from the abnormal sample library to perform incremental training on the multi-label probabilistic classification model, simultaneously optimizing the internal correction parameters of the operating condition-parameter mapping strategy. This structure overcomes the shortcomings of existing technologies, such as fixed models, lack of online iterative optimization mechanisms, and the need for manual parameter recalibration. It continuously updates the identification model and parameter adjustment rules based on actual on-site operating data from the power plant, eliminating the need for manual intervention to complete model adaptation and optimization. It is adaptable to various power plant scenarios, including centralized ground-mounted photovoltaics and distributed industrial and commercial photovoltaics, reducing the workload of multi-scenario engineering implementation and commissioning.

[0025] 6. This invention features a visual interactive module that integrates a working condition display panel, an early warning historical log window, and multi-mode switching buttons. It also includes interactive functions such as view freeze, local waveform magnification, and batch export of test cases. It can simultaneously display photovoltaic power curves, working condition confidence curves, extreme disaster time-series curves, and real-time risk levels. This structure overcomes the shortcomings of traditional simulation interfaces, such as limited functionality, lack of multi-mode view switching, and lack of tools for refined waveform observation and batch test export. It achieves integrated display of multi-dimensional simulation, identification, and early warning data, supporting maintenance personnel in completing standardized experimental operations such as multi-working-condition comparative simulations and extreme disaster simulations.

[0026] 7. The functional modules of this invention adopt loosely coupled standardized data interaction interfaces, and the modular design allows for the tailoring of functional units as needed. This structure is adaptable to various business scenarios such as photovoltaic power plant planning and design, real-time online operation and maintenance, and power grid dispatch simulation. It can be embedded as an independent simulation engine into photovoltaic digital twins and power plant monitoring platforms, or it can be deployed independently to achieve 24 / 7 unattended automated operation, demonstrating strong scenario adaptability and system scalability. Attached Figure Description

[0027] The accompanying drawings, which are provided to further illustrate this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application.

[0028] Figure 1 This is a system framework diagram of the present invention.

[0029] Figure 2 This is the overall flowchart of the present invention. Detailed Implementation

[0030] The embodiments of this application will now be described in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. Furthermore, the following embodiments and features can be combined with each other unless otherwise specified. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0031] Example 1

[0032] Please see Figure 1 This invention provides a full-scenario intelligent photovoltaic power simulation and risk management system. The core concept of this system lies in constructing a fully closed-loop intelligent photovoltaic power simulation and management system, encompassing multi-source data perception, multi-level confidence-based intelligent operating condition identification, two-level hierarchical adaptive parameter correction, chain-like extreme disaster coupling simulation, multi-dimensional matrix hierarchical risk early warning, and sample self-learning iterative optimization. The system mainly includes a basic simulation module, a data fusion and feature extraction module, an operating condition identification module with multi-level confidence-based hierarchical judgment, a two-level adaptive parameter tuning module, a chain-like disaster coupling simulation module, and an active hierarchical risk early warning module.

[0033] The basic simulation module is used to perform power simulation based on the photovoltaic physical mechanism model and simulation parameters, and output simulated power data.

[0034] In this embodiment, the basic simulation module serves as the underlying computing power carrier of the system. Its simulation function realizes the real-time dynamic adjustment and reloading of external parameters through the visualization interaction module, and has built-in photovoltaic physics calculation logic. Among them, the real-time parameter interaction and dynamic reloading mechanism realizes the real-time and smooth adjustment of multi-dimensional simulation parameters through the visualization interaction module, including: (1) Environmental parameter adjustment: through the slider component in the visualization interaction module, the dynamic setting of irradiance, ambient temperature and cloud cover parameters is realized. The adjustable range of irradiance covers 0 to 1500W / m 2 The adjustable range of ambient temperature covers -30°C to 80°C, and the adjustable range of cloud cover covers 0% to 100%. (2) Component parameter adjustment: The photovoltaic module area and photoelectric conversion efficiency are set through the input box. The adjustable range of photovoltaic module area is 0.1m. 2 up to 1000m 2The settable range of photoelectric conversion efficiency is 5% to 30%, and the input box component has a boundary verification function to ensure that the set parameters are within the effective range of the physical and model. (3) Adjustment of operating parameters: Select the data acquisition sampling rate, data time scale and data source type through the drop-down menu component. The sampling rate can be selected for values ​​including but not limited to 1Hz, 5Hz and 10Hz. When any of the environmental parameters, component parameters or operating parameters are adjusted through the visualization interaction module, the basic simulation module immediately triggers the reload of all parameters of the underlying photovoltaic physical mechanism model to ensure that the simulation power data output by the simulation immediately responds to the parameter changes. At the same time, based on the real-time calculation of the photovoltaic physical mechanism model, the simulation power data is continuously output. The simulation power data is synchronously pushed to the data fusion and feature extraction module, the two-level adaptive parameter tuning module and the chain disaster coupling simulation module to realize the data interconnection of the entire module. In addition, the parameter dynamic reload supported by the basic simulation module also includes: when the two-level adaptive parameter tuning module and the chain disaster coupling simulation module issue parameter adjustment instructions, the corresponding simulation parameters inside the model are updated in real time, and the output power curve is refreshed synchronously to ensure that the simulation results respond to the changes in working conditions and disasters in real time.

[0035] In this embodiment, the photovoltaic physical mechanism model is embedded with high-precision physical calculation rules to ensure that the simulation process strictly follows physical laws. The rules include at least: (1) Dynamic calculation of solar altitude angle: Based on the preset geographical latitude and longitude of the photovoltaic power station and real-time time zone information, the solar altitude angle and azimuth angle at each simulation moment are calculated iteratively as the basis for calculating atmospheric optical path and reference irradiance. (2) Triple attenuation layer calculation: The ideal irradiance is calculated and reduced in sequence by atmospheric attenuation, cloud attenuation and temperature influence. Atmospheric attenuation is modeled based on the atmospheric optical path determined by the solar altitude angle, cloud attenuation is modeled based on cloud cover parameters, and temperature influence is modeled based on ambient temperature and temperature coefficient of photovoltaic module, so as to quantify the power loss of photoelectric conversion caused by atmospheric scattering, cloud shading and module temperature drift. (3) Random illumination fluctuation superposition: On the reference power obtained based on the above attenuation calculation, the periodic fluctuation term used to simulate the illumination change under real scene and the adaptive variance Gaussian white noise term used to simulate the random noise of measurement and system are superimposed. (4) Physical constraints on output power: Non-negative physical constraints are applied to the instantaneous simulated power calculated by the model at each time step. Specifically, if the instantaneous simulated power obtained by a single step is less than zero, the power output value at that time is forcibly set to zero. At the same time, linear interpolation is performed to smooth the power values ​​of a fixed number of adjacent sampling points (e.g., two adjacent sampling points) before and after that time to eliminate the sawtooth jumps in the power curve that may be caused by the single-point truncation operation. This avoids the physical meaningless negative power values ​​introduced by meteorological noise interference, dynamic parameter adjustment or extreme disaster coupled calculation process, and ensures that the simulated power data fully conforms to the objective physical laws of photovoltaic power generation.

[0036] In this embodiment, the basic simulation module generates multi-dimensional and multi-frequency simulated power data at a set sampling rate, specifically including: (1) Multi-dimensional and multi-frequency data generation: generating various data including real-time instantaneous power, multi-channel simulated waveforms, historical power sequences, and power prediction trend lines at regular intervals. (2) Dual data verification: before the data is persistently stored, the system automatically performs integrity verification on the generated data sequence to detect whether there are invalid values ​​(such as NaN or Inf) and verifies whether the data volume reaches the predetermined sufficient standard (for example, the number of historical power sequence points is not less than 100 points). (3) Automated intelligent storage: when the generated data volume accumulates to the preset number of points and the verification passes, the system automatically checks the remaining space of the storage medium (implemented through a cross-platform storage monitoring interface, such as calling the disk usage query function). Only when the data verification passes and the remaining disk space is greater than the preset safety threshold (for example, greater than 500MB) will the current data be integrated to generate a structured data file (such as an Excel format file) for persistent storage. The naming of the data file includes precise timestamp information.

[0037] It should be noted that the system can rely on some historical simulation power data to complete feature extraction and model training, thereby enriching the sample dimensions and recognition adaptability of the operating condition identification. This data processing process corresponds to the offline simulation working mode of the system. The core input data source used by the system when performing real-time closed-loop control and equipment status diagnosis is all real-time power data collected on-site at the site. This operating mode corresponds to the online operation and maintenance working mode of the system.

[0038] The data fusion and feature extraction module is configured with offline simulation mode and online operation and maintenance mode. In offline simulation mode, simulated power data, meteorological data and power plant operation status data are collected. In online operation and maintenance mode, simulated power data, real-time power data, meteorological data and power plant operation status data are collected. In both offline simulation mode and online operation and maintenance mode, features are extracted through sliding window and multi-dimensional time series feature vectors are constructed.

[0039] In this embodiment, the data fusion and feature extraction module distinguishes between offline simulation mode and online operation and maintenance mode when performing data acquisition. In offline simulation mode, the acquired data includes three categories: simulated power data output by the basic simulation module, meteorological data sent by meteorological station equipment or numerical weather prediction interface, and power plant operation status data uploaded by inverter and string monitoring equipment. In online operation and maintenance mode, the acquired data includes four categories: simulated power data output by the basic simulation module, real-time power data uploaded by the power plant SCADA system / power acquisition device, meteorological data sent by meteorological station equipment or numerical weather prediction interface, and power plant operation status data uploaded by inverter and string monitoring equipment. Subsequently, all data acquired in both modes are uniformly processed with timestamp alignment to eliminate time-series offset errors between multi-source data.

[0040] In this embodiment, the sliding window is adaptively configured so that the window length and sliding step size are not fixed, but are adaptively adjusted according to the installed capacity of the power station and the frequency of on-site data acquisition. Large-capacity centralized photovoltaic power stations are configured with longer window lengths and larger sliding step sizes, while distributed small-capacity photovoltaic power stations are configured with shorter windows and smaller step sizes, balancing feature recognition accuracy and computing power consumption.

[0041] In this embodiment, the multidimensional time-series feature vector is constructed in layers, including three types of features: (1) Power features: power time-domain features taken from simulated power data or real-time power data on site, including power mean, standard deviation, peak factor, skewness coefficient and slope mean within the window, respectively characterizing average output, fluctuation intensity, instantaneous power mutation, power distribution asymmetry and power rise and fall trend, accurately identifying common operating conditions such as shading, dust accumulation, and cloud fluctuation. (2) Meteorological features: meteorological features taken from meteorological data, calculating window mean, maximum value, minimum value, time-series change rate and trend inflection point, capturing extreme weather and complex weather change features. (3) Status labels of power plant operation status data: discrete status markers such as equipment fault alarm, inverter start and stop, and string disconnection, as auxiliary classification features for operating condition identification.

[0042] In this embodiment, the status labels of power characteristics, meteorological characteristics, and power plant operating status data have significantly different dimensions, requiring unified standardization preprocessing before concatenation: Z-score standardization is used to convert each feature into a dimensionless form using the feature mean and standard deviation within a single sliding window, eliminating dimensional differences; after normalization, the status labels of power characteristics, meteorological characteristics, and power plant operating status data are concatenated sequentially to generate a fixed-dimensional multi-dimensional time-series feature vector. All features are then normalized and concatenated into a fixed-dimensional multi-dimensional time-series feature vector, which is then pushed in batches to the operating condition identification module with multi-level confidence stratification.

[0043] The working condition identification module with multi-level confidence stratification is used to input multi-dimensional time-series feature vectors into a multi-label probabilistic classification model and output the confidence scores of various working conditions. It combines multi-level threshold anti-shake logic with multiple confidence thresholds, differentiated continuous judgment periods, and mutual exclusion priorities of working conditions to determine the working conditions. The working condition types include single normal (such as sunny day, uniform dust accumulation), normal compound (such as dust accumulation + partial cloud cover), single extreme (such as high temperature, sandstorm), chain compound disaster (such as sandstorm + rainfall), normal-extreme cross compound (such as local cloud cover + instantaneous strong wind), and unknown working conditions.

[0044] In this embodiment, the working condition identification module with multi-level confidence stratification is the core module of the present invention. It is used to output the confidence level corresponding to each working condition through a multi-label probability classification model, and to complete the determination of all types of working conditions by relying on multi-level threshold anti-shake logic. This overcomes the technical defects of existing working condition identification logic being simple, prone to misjudgment, and unable to distinguish multiple superimposed composite working conditions.

[0045] In this embodiment, the multi-label probabilistic classification model takes a multi-dimensional temporal feature vector as input and outputs a confidence value corresponding to each working condition label. The multi-label probabilistic classification model selects any one of the machine learning algorithms that can output probabilistic confidence values, such as random forest, XGBoost, and K-nearest neighbors. Each working condition label is independent of the others, and high-confidence judgment results for multiple working conditions can be output simultaneously, meeting the recognition requirements of conventional composite and conventional-extreme cross-composite working conditions. The offline model training dataset consists of a mixture of 60% system simulation labeled samples and 40% power plant field measured labeled samples. The sample labeling rule is to match a unique main working condition label in a single window temporal sequence, and to simultaneously label multiple types of labels for composite working conditions. Incremental training adopts an elastic weight update strategy, and the loss weight of newly added unknown samples is set to 1.5 times the weight of conventional historical samples, giving priority to adapting to new abnormal working condition distributions. The number of training iterations is fixed at 20 rounds, and the iteration convergence threshold is set to 0.001.

[0046] In this embodiment, the multi-level threshold anti-shake logic is configured with four types of confidence judgment thresholds, six types of continuous judgment periods dedicated to different working conditions, and a preset mutual exclusion priority order for valid working condition judgments. The specific configuration is as follows.

[0047] 1. The four confidence thresholds include a regular high confidence threshold, an extreme high confidence threshold, a general medium confidence threshold, and an unknown low confidence threshold. The values ​​of the four thresholds satisfy the following conditions: the regular high confidence threshold is greater than the extreme high confidence threshold, the extreme high confidence threshold is greater than the general medium confidence threshold, and the general medium confidence threshold is greater than the unknown low confidence threshold. Setting a regular high confidence threshold can filter out false identifications caused by instantaneous cloud disturbances, reducing the extreme high confidence threshold can shorten the identification delay of extreme disasters, the general medium confidence threshold is used for the determination of complex working conditions, and the unknown low confidence threshold serves as the critical value for the determination of unknown working conditions.

[0048] 2. The six types of continuous judgment cycles are: single normal judgment cycle, normal composite judgment cycle, normal-extreme cross-composite judgment cycle, single extreme judgment cycle, chain-type composite disaster judgment cycle, and unknown condition judgment cycle. The duration relationship of the first five types of continuous judgment cycles is as follows: single normal judgment cycle is longer than normal composite judgment cycle, normal composite judgment cycle is longer than normal-extreme cross-composite judgment cycle, normal-extreme cross-composite judgment cycle is longer than single extreme judgment cycle, and the duration of single extreme judgment cycle is equal to that of chain-type composite disaster judgment cycle. It should be noted that the unknown condition judgment cycle is only used for fallback identification when all judgments of the first five types of conditions are invalid. It does not participate in the synchronous mutual exclusion judgment of the first five types of conditions and adopts independent duration parameter configuration. It does not need to participate in the duration comparison relationship of the aforementioned cycles. A longer judgment cycle is configured for normal conditions to achieve anti-shaking suppression, and a shortest judgment cycle is configured for single extreme conditions or chain-type composite disaster conditions to achieve rapid triggering of extreme disaster simulation and risk warning.

[0049] 3. The priority of valid operating conditions is determined from high to low as follows: chain-like composite disaster, single extreme disaster, normal-extreme cross-composite disaster, normal composite disaster, and single normal disaster. If any high-priority valid operating condition meets the corresponding judgment condition, the judgment process of all low-priority valid operating conditions will be blocked to avoid conflicts in the simultaneous judgment of multiple operating conditions. The unknown operating condition judgment process will only be executed when all valid operating conditions do not meet their respective judgment conditions.

[0050] The judgment rules for each working condition include: (1) Single normal working condition judgment rule: If the confidence level of any normal working condition label is continuously higher than the normal high confidence threshold within the complete single normal judgment period, the current working condition is judged as a single normal working condition; (2) Normal compound working condition judgment rule: If no normal working condition label meets the single working condition judgment condition corresponding to the normal high confidence threshold, and the confidence level of two or more normal working condition labels is continuously higher than the general medium confidence threshold within the complete normal compound judgment period, the current working condition is judged as a normal compound working condition; (3) Single extreme working condition judgment rule: If the confidence level of any single extreme working condition label is continuously higher than the extreme high confidence threshold within the complete single extreme judgment period, the current working condition is judged as a single extreme working condition, and the subsequent judgment process for all normal working conditions is terminated; (4) Chain compound disaster working condition judgment rule: If no single extreme working condition label meets the single working condition judgment condition corresponding to the extreme high confidence threshold, and the confidence level of two or more single extreme working condition labels is continuously higher than the general medium confidence threshold within the complete normal compound judgment period, the current working condition is judged as a single extreme working condition, and the subsequent judgment process for all normal working conditions is terminated; If the confidence level of the end-type working condition label is continuously higher than the general medium confidence threshold during the complete chain-type composite disaster judgment period, then the current working condition is determined to be a chain-type composite disaster working condition, and the judgment process of all low-priority valid working conditions is terminated; (5) Normal-extreme cross-composite working condition judgment rule: If there is no normal-type working condition label that meets the single working condition judgment condition corresponding to the normal high confidence threshold, and there is no single extreme-type working condition label that meets the single working condition judgment condition corresponding to the extreme high confidence threshold, and the confidence level of at least one type of normal-type working condition label and at least one type of single extreme-type working condition label is continuously higher than the general medium confidence threshold during the complete normal-extreme cross-composite judgment period, then the current working condition is determined to be a normal-extreme cross-composite working condition; (6) Unknown working condition judgment rule: If the confidence level of all normal-type, single extreme-type, and composite-type working condition labels is continuously lower than the unknown low confidence threshold during the complete unknown working condition judgment period, then the current working condition is determined to be an unknown working condition, and the original operating data and multi-dimensional time series feature vector corresponding to the current sliding window are archived and stored in the abnormal sample library. In addition, the system temporarily stores unidentified operating conditions in an abnormal sample database. Maintenance personnel can query these samples through the monitoring interface and manually label (or confirm) their true operating conditions based on historical operation logs, meteorological data, alarm information, etc. The system then imports the labeled samples into the incremental training process to update the model.

[0051] The working condition identification module with multi-level confidence stratification will synchronously transmit the final working condition and the confidence level corresponding to each working condition label to the two-level adaptive parameter tuning module, the chain disaster coupling simulation module and the visualization interaction module.

[0052] The two-level adaptive parameter tuning module is used to adaptively adjust the simulation parameters according to the working condition identification results. This includes coarse gear adjustment through the working condition-parameter mapping strategy, and fine-tuning correction of PID with parameter limiting constraints based on the deviation between the simulated power data and the real-time power data on site.

[0053] In this embodiment, the two-level adaptive parameter tuning module receives the working condition judgment result output by the working condition identification module with multi-level confidence layer judgment, and completes the automatic correction of simulation parameters in two levels. The two-level correction process is equipped with a smooth transition buffer and a dual stability constraint mechanism of parameter range limiting to avoid the technical problems of simulation model distortion caused by instantaneous parameter jumps and parameter over-limits.

[0054] 1. Operating Condition-Parameter Mapping Coarse Adjustment Strategy: This module has a pre-configured operating condition-parameter mapping strategy. This strategy stores the irradiance attenuation coefficient and photoelectric conversion efficiency correction levels corresponding to various common operating scenarios such as dust accumulation, partial shading, and component aging. When the operating condition identification module outputs a stable operating condition judgment result, the module matches the parameter level of the corresponding operating condition within the mapping strategy to complete the first-level large-span simulation parameter correction. A smooth transition buffer is set during the level switching process, and the simulation parameters are updated gradually in multiple stages along the buffer to eliminate the instantaneous jump phenomenon of the simulation power time series curve caused by parameter mutations.

[0055] 2. PID Fine-Tuning Correction Logic: After completing the coarse gear adjustment process, the relative error between the simulated power data and the real-time field power data is calculated in real time. Only when the relative error exceeds the preset relative error threshold for two consecutive acquisition cycles will the PID controller be driven to perform fine-tuning of the control parameters in small steps. For example, if the system data acquisition cycle is set to 1 second and the preset relative error threshold is 5%, after the coarse gear adjustment is completed, the simulated power data of 1000kW is used as the benchmark for deviation monitoring. When the real-time field power data in the first acquisition cycle is 935kW with a relative error of 6.5%, and the real-time field power data in the second acquisition cycle is 930kW with a relative error of 7.0%, and the relative error values ​​in two consecutive acquisition cycles both exceed the preset relative error threshold of 5%, the system immediately drives the PID controller to perform fine-tuning of the control parameters according to the preset small step gradient. If only a single cycle of error exceeds the limit and the error returns to the relative error threshold range in the next cycle, it is determined to be an instantaneous operating condition disturbance. The PID controller remains in standby mode and does not output a correction command to avoid frequent control loop actions that could cause system oscillation.

[0056] All adjustable simulation parameters are configured with preset fixed value ranges as amplitude constraints. During PID correction, the parameter values ​​must not exceed the upper and lower limits of the corresponding range to avoid the simulation model violating the inherent physical characteristics of photovoltaic power generation due to parameter over-limitation.

[0057] The two-level adaptive parameter tuning module continuously transmits the updated simulation parameters to the basic simulation module, so that the simulation power timing output by the basic simulation module matches the actual power output and operation characteristics of the photovoltaic power station on site.

[0058] The chain disaster coupling simulation module is used to receive the judgment results of a single extreme working condition or a chain compound disaster working condition, retrieve historical measured data of extreme weather that matches the current received working condition, extract extreme time series parameters using Monte Carlo random sampling, and perform coupled calculations using disaster weighting factors and time decay functions. The results are then input into the basic simulation module to deduce the power impact of the chain disaster. The extreme time series parameters are a meteorological dynamic subset of the simulated power data.

[0059] In this embodiment, the chain disaster coupling simulation module is used to receive the judgment results including single extreme working conditions or chain compound disaster working conditions, and to carry out time-series disaster power extrapolation for these two types of working conditions, overcoming the technical limitations of the prior art that can only independently simulate single extreme weather and cannot quantify the secondary attenuation effects caused by the chain superposition of multiple disasters.

[0060] In this embodiment, the extreme time-series parameter generation process specifically includes: the chain disaster coupling simulation module receives the judgment results of a single extreme working condition or a chain composite disaster working condition output by the working condition identification module. The aforementioned working condition judgment results are obtained through multi-level confidence hierarchical judgment logic calculation; the module retrieves the historical measured database of extreme weather that matches the current effective working condition, and fits the statistical distribution law corresponding to various meteorological elements such as wind speed, dust concentration, rainfall, and temperature. Temperature and wind speed follow a normal distribution, while dust concentration and rainfall follow a log-normal distribution; based on the above statistical distribution model, the Monte Carlo random sampling algorithm is used to generate extreme time-series parameters that run through the entire simulation duration and whose values ​​change dynamically over time. The extreme time-series parameters are a meteorological dynamic subset constituting the simulation power data, including extreme time-series parameters such as light intensity, ambient temperature, ambient wind speed, dust concentration, and rainfall intensity.

[0061] In this embodiment, the disaster coupling calculation rules specifically include: the chain-type disaster coupling simulation module configures a disaster dynamic weighting factor and a time decay function, relying on the two to complete the multi-disaster superposition coupling calculation; the disaster dynamic weighting factor increases with the number of disaster types simultaneously superimposed. The more disasters superimposed, the larger the correction weight value corresponding to irradiance decay and photovoltaic module decay. For example, the disaster dynamic weighting factor is configured as follows: if the rainfall event occurs within 3 hours after the sandstorm ends, it is determined to be a high mud adhesion scenario, and the corresponding high weighting coefficient is activated; if the sandstorm ends more than 3 hours later, a low weighting factor is activated. The weighting coefficient and the time decay function are piecewise linear functions that monotonically decrease with the simulation duration. The time decay function characterizes the physical process of the gradual dissipation of secondary decay effects from disasters such as dust accumulation and mud accumulation over time. The piecewise calculation formula is as follows: when the simulation duration t satisfies 0h < t ≤ 6h, the decay correction coefficient K(t) = 1 - 0.08t; when the simulation duration t satisfies 6h < t ≤ 24h, the decay correction coefficient K(t) = 0.52 - 0.02(t - 6); when t > 24h, K(t) is fixed at 0.04, and the secondary decay effect is basically dissipated. This module reads the simulation step duration in real time, substitutes it into the piecewise function to solve for the decay correction coefficient at the current moment, and multiplies it by the disaster dynamic weighting factor to obtain the total decay correction weight.

[0062] It should be noted that the normal-extreme cross-combination working condition only triggers the parameter correction process of the two-level adaptive parameter tuning module, and does not execute the extreme time-series sampling and multi-hazard coupling simulation calculation of the chain disaster coupling simulation module.

[0063] In this embodiment, the chain disaster power impact simulation process specifically includes: inputting all extreme time-series parameters after coupling calculation correction into the basic simulation module, driving the basic simulation module to output the chain disaster full-cycle power time-series curve; the module statistically analyzes the two core evaluation indicators, the maximum power loss rate and the cumulative power outage duration, within the complete simulation cycle, and transmits the obtained chain disaster power simulation evaluation results to the active graded risk warning module.

[0064] The proactive risk warning module is used to retrieve simulation results of the power impact of chain disasters, match them with a multi-dimensional risk matrix, and automatically generate standardized risk warning information.

[0065] In this embodiment, the multidimensional risk matrix uses power loss rate and expected power outage duration as its two core evaluation dimensions. The multidimensional risk matrix has five preset risk levels: safety, caution, early warning, warning, and emergency. Each risk level is associated with a corresponding equipment safety risk description, standardized early warning content, and tiered operation and maintenance guidelines. The equipment safety risk description characterizes the type of safety hazard corresponding to the photovoltaic modules and electrical equipment at the current level. The standardized early warning content is a unified alarm text containing quantitative indicators of operating conditions, disasters, and power loss. The tiered operation and maintenance guidelines are standardized operation and maintenance strategies matched to the risk level. The tiered early warning information generated by this module is synchronously transmitted to the visual interaction module for pop-up display and written to the early warning history log for persistent storage.

[0066] This system also includes a self-learning iteration module, which periodically reads archived data (i.e., raw data of unknown operating conditions and multi-dimensional time-series feature vectors) from the abnormal sample library. After confirming or supplementing the operating condition labels of the archived data, it updates the internal weight parameters of the multi-label probabilistic classification model using incremental training. At the same time, based on the actual power deviation corresponding to the sample features, it optimizes the correction gear values ​​of each scenario within the operating condition-parameter mapping strategy, so that the accuracy of operating condition identification and parameter correction continuously improves as the system runs, thus constructing a complete self-evolving closed loop.

[0067] The system also includes a visualization interaction module, which can provide an integrated human-computer interaction interface for operation and maintenance personnel. It includes three fixed panels and multiple interactive operation functions: (1) Operating condition display panel: displays the photovoltaic power curve, operating condition confidence change curve, and extreme disaster time series curve in real time, and displays the current identified operating condition, corresponding confidence level, the parameter adjustment mode being executed by the system, and the real-time risk level simultaneously; (2) Early warning history log window: stores all historical graded early warning events, and supports retrieval and viewing by time, risk level, and disaster type; (3) Multi-mode switching button: supports switching between two operating views: offline simulation working mode and online operation and maintenance working mode. Different modes display the corresponding data charts and function control panels differently; (4) Interactive operation function: built-in view freeze, local waveform magnification, and batch export of simulation test cases, which facilitates operation and maintenance personnel to carry out extreme disaster pre-drills and multi-operating condition comparison simulation tests.

[0068] The following detailed description of the implementation of this invention, using a complete system architecture and actual photovoltaic power plant operation examples, is provided. These examples are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0069] This embodiment is applied to a 100MW centralized ground-mounted photovoltaic power station in a plain. It is equipped with a local meteorological monitoring station and an inverter SCADA data acquisition device. The full-scenario photovoltaic power intelligent simulation and risk management system of this invention is deployed. The hardware carrier is an industrial server, the software environment is a Linux operating system, and the multi-label probabilistic classification model uses the XGBoost gradient boosting tree algorithm.

[0070] 1. System initialization and deployment: Deploy this system on the industrial server supporting the 100MW centralized ground photovoltaic power station. During the system initialization phase, complete the parameter configuration of each module in sequence.

[0071] The basic simulation module loads the photovoltaic physical mechanism model, inputs the fixed basic parameters of the power station, including the total area of ​​the modules, the photoelectric conversion efficiency of a single module, the geographical latitude and longitude of the site, and the altitude, and defines the initial value range of each simulation parameter.

[0072] The data fusion and feature extraction module completes the multi-source data interface docking, synchronously accessing real-time meteorological time series data from the meteorological station, real-time power data from the inverter and string, and simulated power data output by the basic simulation module; based on the 100MW large-capacity installed capacity, the sliding window parameters are adaptively configured, with a sliding window length of 30 sampling points and a sliding step size of 5 sampling points, and the system's unified sampling frequency is set to 1Hz.

[0073] The working condition identification module with multi-level confidence stratification completes the configuration of threshold, judgment period, and judgment priority parameters: the normal high confidence threshold is 0.75, the extreme high confidence threshold is 0.65, the general medium confidence threshold is 0.6, and the unknown low confidence threshold is 0.4; the single normal judgment period is 5 sliding windows, the normal composite judgment period is 3 sliding windows, the normal-extreme cross-composite judgment period is 2 sliding windows, and the single extreme judgment period and the chain composite disaster judgment period are both 1 sliding window; the effective working condition judgment priority is fixed from high to low as chain composite disaster, single extreme, normal-extreme cross-composite, normal composite, and single normal.

[0074] The two-level adaptive parameter adjustment module loads the working condition-parameter mapping level table, which pre-stores the irradiance attenuation correction value and photoelectric conversion efficiency correction level corresponding to four common working conditions: light dust accumulation, heavy dust accumulation, partial shading, and uniform aging. The smooth transition buffer between levels is set to 30 seconds. The PID fine-tuning start error threshold is set to 3%. The adjustable parameter limit range is limited to: photoelectric conversion efficiency 0.05~0.3, irradiance attenuation coefficient 0.2~1.0.

[0075] The chain-type disaster coupling simulation module accesses the local historical measured database of extreme weather over the past ten years, and records the statistical distribution patterns of wind speed, ambient temperature, and dust concentration corresponding to four types of extreme weather: high temperature, strong wind, dust storm, and rainstorm. The configuration rules for the disaster dynamic weighting factor are as follows: the weight for superimposing a single disaster is 1.0, the weight for superimposing two types of disasters is 1.2, and the weight for superimposing three types of disasters is 1.4. The time decay function adopts a piecewise monotonically decreasing form, with rapid decay calculation performed in the 0-6h interval and slow decay calculation performed in the 6-24h interval.

[0076] The proactive risk warning module has a built-in multi-dimensional risk matrix with five preset risk judgment standards: a power loss rate of no more than 10% corresponds to the blue alert level, a power loss rate of more than 10% but no more than 30% corresponds to the yellow warning level, a power loss rate of more than 30% but no more than 50% corresponds to the orange warning level, and a power loss rate of more than 50% corresponds to the red emergency level.

[0077] The self-learning iteration module is configured with a scheduled task to automatically read archived data from the abnormal sample library at 2 AM every day to complete incremental training of the multi-label probabilistic classification model; the visualization interaction module is deployed on a local workstation client to support parallel access from multiple terminals.

[0078] 2. The implementation process of automatic identification and adaptive parameter adjustment under normal working conditions includes the following steps.

[0079] After 72 hours of continuous operation under clear skies, light dust accumulated on the surface of the photovoltaic modules, and the real-time power data on site remained lower than the benchmark simulation power output by the basic simulation module based on the initial fixed parameters.

[0080] The data fusion and feature extraction module updates a set of multi-dimensional time-series feature vectors corresponding to a complete sliding window every 5 seconds. The mean index of the power feature continues to decline, while the standard deviation index rises slightly.

[0081] The working condition identification module with multi-level confidence stratification outputs a confidence value of 0.81 for the light dust accumulation working condition. This confidence value is higher than the conventional high confidence threshold of 0.75 for five consecutive complete single normal judgment cycles, and there is no higher priority valid working condition trigger. The system determines that the current operating condition is a light dust accumulation single normal working condition.

[0082] The two-level adaptive parameter adjustment module retrieves the operating condition-parameter mapping table to match the photoelectric conversion efficiency correction level corresponding to light dust accumulation. It then adjusts the photoelectric conversion efficiency parameter in multiple steps along a 30s smooth buffer to complete the first level coarse adjustment. After the coarse adjustment, it continuously calculates the relative error between the simulated power and the real-time power data. When the relative error stabilizes at 4.2% and exceeds the 3% relative error threshold for three consecutive acquisition cycles, it starts the PID controller to perform small-step fine correction. Finally, the relative error between the simulated power data and the real-time power data converges to 2.1%, and all adjustable parameters do not exceed the preset limit range throughout the process.

[0083] The basic simulation module continuously outputs simulated power time-series curves that closely match the actual power output characteristics of the power station; the visual interactive module's operating condition display panel synchronously displays the current judgment operating condition, 0.81 operating condition confidence level, automatic parameter adjustment operation mode, green safety risk level, and generates and pushes graded early warning information.

[0084] 3. The implementation process of chain-type composite disaster coupling simulation and graded early warning includes the following steps.

[0085] The local meteorological monitoring center issued an extreme weather warning for the superimposed high temperature and sandstorm, and the operators switched the visual interactive module to extreme test mode.

[0086] The working condition identification module with multi-level confidence stratification outputs a confidence level of 0.68 for high temperature working condition and 0.71 for high dust working condition within a complete single extreme judgment cycle. Both values ​​are higher than the extreme high confidence threshold of 0.65. Based on the preset working condition judgment priority, the two extreme working conditions simultaneously meet the general medium confidence threshold judgment condition. The system determines that the current working condition is a high temperature-dust sandstorm chain composite disaster working condition and terminates all low-priority valid working condition judgment processes.

[0087] The chain disaster coupling simulation module retrieves historical statistical distributions of extreme weather events such as high temperature and dust storms, and uses a Monte Carlo random sampling algorithm to generate time-series dynamic parameters of temperature and dust concentration covering the next 12 hours of simulation. The dynamic weighting coefficient of 1.2 corresponding to the two types of disasters is superimposed, and a piecewise time decay function is introduced to characterize the physical law of the gradual dissipation of dust adhesion decay effect over time. The coupled and corrected extreme time-series parameters are input into the basic simulation module to deduce the complete 12-hour chain disaster power time-series curve, and two quantitative evaluation indicators are obtained: the maximum power loss rate of 58% within the cycle and the expected cumulative power outage duration of 9 hours.

[0088] The proactive risk warning module retrieves a multi-dimensional risk matrix to complete indicator matching. When the power loss rate is 58% > 50%, it matches the red emergency risk level and generates standardized graded warning information. The warning information records the warning level, the triggering of a chain-like high-temperature and dust storm, the estimated peak power loss of 58%, the high safety risk of component wear and panel flashover tripping, and the corresponding emergency shutdown, windproof and dustproof operation and maintenance guidelines.

[0089] Standardized red alert information is transmitted to a pop-up window of the visualization interaction module and persistently written to the alert history log window. Operation and maintenance personnel can view the disaster time series curve and power loss quantification prediction data based on the extreme test view, and export the chain disaster simulation test cases in batches for archiving and retention.

[0090] 4. The process of archiving and implementing self-learning iterations for unknown working condition samples includes the following steps.

[0091] The power plant encountered continuous rainy weather combined with a local component short-circuit fault. The power waveform characteristics lacked matching known operating condition labels, and the confidence levels of all normal and extreme labels remained below the unknown low-confidence threshold of 0.4 for five consecutive windows. The system determined the current operating condition to be unknown and archived the original data (including original power and weather data) and multi-dimensional feature vectors corresponding to the current sliding window into the anomaly sample library. Every day at 2:00 AM, the self-learning iteration module automatically reads the archived unknown operating condition samples from the anomaly sample library. After manually supplementing the true operating condition labels, incremental training is performed on the XGBoost multi-label probabilistic classification model, simultaneously iteratively optimizing the parameter correction level corresponding to the short-circuit fault within the operating condition-parameter mapping strategy. When the power plant experiences a similar component short-circuit fault again, the system can accurately identify the fault condition and automatically match the attenuation correction parameters to complete the adaptive correction of the simulation model.

[0092] By leveraging the synergistic effects of the above modules, a comprehensive intelligent simulation and risk management system for photovoltaic power across all scenarios has been constructed. This system features high-confidence probability identification, multi-level adaptive parameter tuning, dynamic coupling of chain disasters, and proactive hierarchical early warning capabilities. It successfully addresses the core pain points of traditional simulation technologies, such as coarse condition identification, poor parameter adaptability, lack of extreme disaster simulation capabilities, and the disconnect between simulation results and risk management. It upgrades the photovoltaic power simulation model from isolated fixed models, manual experience-based parameter tuning, and offline single-scenario extrapolation to a fully intelligent closed-loop management paradigm driven by multi-source real-time data. This paradigm includes autonomous diagnosis of operating conditions and precise self-calibration of model parameters, simulation of the dynamic impact of chain disasters across cycles, and quantitative results directly driving structured proactive early warning. It achieves full-process integration from data fusion perception, intelligent condition identification and self-learning, refined adaptive model simulation to risk quantification assessment and precise handling suggestions. This system can stably, autonomously, and accurately respond to all scenarios of photovoltaic power plants, from standard operation and composite performance degradation to chain extreme disasters, effectively improving power prediction accuracy, equipment health diagnosis capabilities, and the system's proactive defense level.

[0093] Example 2

[0094] Please see Figure 2 Applied to the system described in Example 1, the present invention also provides a method for intelligent simulation and risk management of photovoltaic power in all scenarios, which specifically includes the following steps.

[0095] S1. Power simulation is performed based on the photovoltaic physical mechanism model and simulation parameters through the basic simulation module, and the simulated power data is output.

[0096] S2. Configure offline simulation working mode and online operation and maintenance working mode through data fusion and feature extraction module; in offline simulation working mode, collect simulation power data, meteorological data and power plant operation status data; in online operation and maintenance working mode, collect simulation power data, on-site real-time power data, meteorological data and power plant operation status data; in both offline simulation working mode and online operation and maintenance working mode, extract features through sliding window and construct multi-dimensional time series feature vector.

[0097] S3. The working condition identification module with multi-level confidence stratification judgment inputs the multi-dimensional time series feature vector into the multi-label probability classification model and outputs the confidence level of various working conditions. Combined with the multi-level threshold anti-shaking logic with multiple confidence thresholds, differentiated continuous judgment period and working condition mutual exclusion priority, the working condition is judged. The working condition types include single normal working condition, normal compound working condition, single extreme working condition, chain compound disaster working condition, normal-extreme cross compound working condition and unknown working condition.

[0098] S4. The simulation parameters are adaptively adjusted based on the working condition identification results through a two-level adaptive parameter adjustment module. This includes coarse adjustment of gears through a working condition-parameter mapping strategy, and fine adjustment of PID with parameter limiting constraints based on the deviation between the simulated power data and the real-time power data on site.

[0099] S5. The chain disaster coupling simulation module receives the judgment results of a single extreme working condition or a chain compound disaster working condition, retrieves the historical measured data of extreme weather that matches the current received working condition, extracts extreme time series parameters by Monte Carlo random sampling, and performs coupled calculations by combining the disaster weighting factor and the time decay function. The results are then input into the basic simulation module to deduce the power impact of the chain disaster. The extreme time series parameters are a meteorological dynamic subset in the simulated power data.

[0100] S6. By retrieving the simulation results of the power impact of chain disasters through the proactive risk warning module, a multi-dimensional risk matrix is ​​matched, and standardized graded warning information is automatically generated.

[0101] By combining the above execution steps, this invention constructs a complete closed loop from data perception, intelligent diagnosis, adaptive simulation, disaster prediction to proactive early warning and self-evolution, providing a comprehensive, accurate and forward-looking solution for the intelligent operation and maintenance and risk management of photovoltaic power plants.

[0102] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A full-scenario photovoltaic power intelligent simulation and risk management system, characterized in that, include: The basic simulation module is used to perform power simulation based on the photovoltaic physical mechanism model and simulation parameters, and output simulated power data. The data fusion and feature extraction module is configured with offline simulation mode and online operation and maintenance mode. In offline simulation mode, it collects simulated power data, meteorological data and power plant operation status data. In online operation and maintenance mode, it collects simulated power data, real-time power data, meteorological data and power plant operation status data. In both offline simulation mode and online operation and maintenance mode, features are extracted through sliding window and multi-dimensional time series feature vectors are constructed. The working condition identification module with multi-level confidence stratification is used to input multi-dimensional time-series feature vectors into a multi-label probability classification model and output the confidence level of various working conditions. It combines multi-level threshold anti-shaking logic with multiple confidence thresholds, differentiated continuous judgment period, and mutual exclusion priority of working conditions to determine the working conditions. The working condition types include single normal working conditions, normal compound working conditions, single extreme working conditions, chain compound disaster working conditions, normal-extreme cross compound working conditions, and unknown working conditions. The two-level adaptive parameter tuning module is used to adaptively adjust the simulation parameters according to the working condition identification results, including coarse gear adjustment through the working condition-parameter mapping strategy, and fine-tuning correction of PID with parameter limiting constraints based on the deviation between the simulated power data and the real-time power data on site. The chain disaster coupling simulation module is used to receive the judgment results of a single extreme working condition or a chain compound disaster working condition, retrieve historical measured data of extreme weather that matches the current received working condition, extract extreme time series parameters using Monte Carlo random sampling, and perform coupled calculations using disaster weighting factors and time decay functions. The results are then input into the basic simulation module to deduce the power impact of the chain disaster. The extreme time series parameters are a meteorological dynamic subset of the simulated power data. The proactive risk warning module is used to retrieve simulation results of the power impact of chain disasters, match them with a multi-dimensional risk matrix, and automatically generate standardized risk warning information.

2. The intelligent simulation and risk management system for full-scenario photovoltaic power according to claim 1, characterized in that, In the data fusion and feature extraction module, the step size and window length of the sliding window are adaptively configured according to the installed capacity of the power plant and the data acquisition frequency. The multidimensional time-series feature vector includes: Power time-domain features derived from simulated power data or real-time power data from the field, including power mean, standard deviation, peak factor, skewness coefficient, and mean slope; Meteorological features derived from meteorological data include the mean, maximum, minimum, rate of change, and trend inflection point of irradiance, temperature, wind speed, and dust concentration within the same sliding window; The status labels are taken from the power plant's operating status data. The status labels are discrete status markers for equipment fault alarms, inverter start-up and shutdown, and string disconnection from the grid.

3. The intelligent simulation and risk management system for full-scenario photovoltaic power according to claim 1, characterized in that, The multi-level threshold anti-shake logic is configured with four types of confidence thresholds, namely, regular high confidence threshold, extreme high confidence threshold, general medium confidence threshold, and unknown low confidence threshold. The numerical relationship of the four types of confidence thresholds is as follows: regular high confidence threshold is greater than extreme high confidence threshold, extreme high confidence threshold is greater than general medium confidence threshold, and general medium confidence threshold is greater than unknown low confidence threshold. The multi-level threshold anti-shake logic is configured with six types of continuous judgment cycles dedicated to different working conditions: single normal judgment cycle, normal composite judgment cycle, single extreme judgment cycle, chain-type composite disaster judgment cycle, normal-extreme cross-composite judgment cycle, and unknown working condition judgment cycle. The duration of the first five types of continuous judgment cycles is as follows: single normal judgment cycle is longer than normal composite judgment cycle, normal composite judgment cycle is longer than normal-extreme cross-composite judgment cycle, normal-extreme cross-composite judgment cycle is longer than single extreme judgment cycle, and the duration of single extreme judgment cycle is equal to that of chain-type composite disaster judgment cycle. The unknown working condition judgment cycle is configured with an independent duration and is not included in the duration ranking comparison of the first five types of working conditions. The multi-level threshold anti-shake logic presets the priority order of the working condition judgment, from high to low as follows: chain-like composite disaster, single extreme, normal-extreme cross-composite, normal composite, and single normal. If the high-priority effective working condition judgment condition is met, the judgment process of all low-priority effective working conditions is blocked. The unknown working condition judgment process is only executed when all effective working condition judgment conditions are not met.

4. The intelligent simulation and risk management system for full-scenario photovoltaic power according to claim 3, characterized in that, The multi-level threshold anti-shake logic is configured as follows: If, without triggering a higher priority valid working condition, the confidence level of any regular working condition label is continuously higher than the regular high confidence threshold within a single regular judgment period, it is judged as a single regular working condition. If no regular working condition meets the high-confidence single working condition condition of the regular type, and the confidence level of two or more regular working condition labels is consistently higher than the general medium confidence threshold within the regular composite judgment period, it is judged as a regular composite working condition. If the confidence level of any single extreme type of working condition label is continuously higher than the extreme high confidence threshold within the single extreme judgment period, it is judged as a single extreme working condition, and all subsequent judgments of regular types of working conditions are terminated. If no single extreme type of working condition meets the high-confidence single working condition condition for extreme type, and the confidence level of two or more single extreme type of working condition labels is continuously higher than the general medium confidence threshold within the chain composite disaster judgment period, it is judged as a chain composite disaster working condition, and all low-priority valid working condition judgments are terminated. If no corresponding high-confidence single-condition triggers are found in either the normal or extreme operating conditions, and the confidence level of at least one normal condition and one extreme condition label is consistently higher than the general medium-confidence threshold within the normal-extreme cross-composite determination period, then it is determined to be a normal-extreme cross-composite operating condition. If the confidence level of all known operating condition labels is lower than the unknown low confidence threshold throughout the unknown operating condition determination period, it is determined to be an unknown operating condition, and the corresponding original data and multi-dimensional feature vectors are automatically archived to the abnormal sample library.

5. The intelligent simulation and risk management system for full-scenario photovoltaic power according to claim 1, characterized in that, The pre-stored working condition-parameter mapping strategy includes irradiance attenuation and photoelectric conversion efficiency correction levels corresponding to dust accumulation, shading, and aging scenarios. A smooth transition buffer is set during level switching to avoid abrupt changes in simulation parameters. The deviation judgment logic of the PID fine-tuning correction is as follows: when the relative error between the simulated power and the real-time power data on site exceeds the set threshold for two consecutive acquisition cycles, the PID controller is started to perform small-step parameter correction. The parameter amplitude constraint is a preset fixed value range.

6. The intelligent simulation and risk management system for full-scenario photovoltaic power according to claim 1, characterized in that, The disaster weighting factor increases dynamically with the number of disasters superimposed, and the time decay function is a piecewise function that decreases monotonically with the simulation duration, used to characterize the gradual dissipation of the secondary impact of disasters over time.

7. The intelligent simulation and risk management system for full-scenario photovoltaic power according to claim 1, characterized in that, The multidimensional risk matrix uses power loss rate and expected power outage duration as its two core dimensions, and matches them with corresponding early warning levels and graded operation and maintenance guidelines.

8. The intelligent simulation and risk management system for full-scenario photovoltaic power according to claim 1, characterized in that, It also includes a self-learning iteration module, which is used to periodically read archived data in the abnormal sample library. After confirming or supplementing the working condition labels of the archived data, it incrementally trains the multi-label probabilistic classification model and synchronously updates the gear parameters of the working condition-parameter mapping strategy.

9. The intelligent simulation and risk management system for full-scenario photovoltaic power according to claim 1, characterized in that, It also includes a visualization interaction module, which provides a working condition display panel, an early warning history log window, and a multi-mode switching button, and is configured with interactive operation functions such as view freeze, local waveform magnification, and batch export of test cases.

10. A method for intelligent simulation and risk management of photovoltaic power across all scenarios, characterized in that, include: Power simulation is performed based on the photovoltaic physical mechanism model and simulation parameters through the basic simulation module, and the simulated power data is output. The data fusion and feature extraction module configures offline simulation mode and online operation and maintenance mode. In offline simulation mode, simulated power data, meteorological data and power plant operation status data are collected. In online operation and maintenance mode, simulated power data, real-time power data, meteorological data and power plant operation status data are collected. In both offline simulation mode and online operation and maintenance mode, features are extracted through sliding window and multi-dimensional time series feature vectors are constructed. The working condition identification module with multi-level confidence stratification inputs multi-dimensional time-series feature vectors into a multi-label probability classification model and outputs the confidence scores of various working conditions. The working condition is determined by multi-level threshold anti-shaking logic with multiple confidence thresholds, differentiated continuous judgment periods, and mutual exclusion priorities of working conditions. The working condition types include single normal working conditions, normal compound working conditions, single extreme working conditions, chain compound disaster working conditions, normal-extreme cross compound working conditions, and unknown working conditions. The simulation parameters are adaptively adjusted based on the working condition identification results through a two-level adaptive parameter tuning module. This includes coarse gear adjustment through a working condition-parameter mapping strategy, and fine-tuning of PID with parameter limiting constraints based on the deviation between the simulated power data and the real-time power data on site. The chain disaster coupling simulation module receives the judgment results of a single extreme working condition or a chain compound disaster working condition, retrieves historical measured data of extreme weather that matches the current received working condition, extracts extreme time series parameters using Monte Carlo random sampling, and performs coupled calculations using disaster weighting factors and time decay functions. The results are then input into the basic simulation module to deduce the power impact of the chain disaster. The extreme time series parameters are a meteorological dynamic subset of the simulated power data. By retrieving simulation results of the power impact of chain disasters through the proactive risk warning module and matching them with a multidimensional risk matrix, standardized risk warning information is automatically generated.

Images