Simulation platform and simulation method for photovoltaic power prediction based on fitting correction and digital twinning

Abstract

This invention discloses a photovoltaic power prediction simulation platform based on fitting correction and digital twins, comprising modules such as a hybrid data source, scheme configuration and comparison, a digital twin simulation engine, scenario scripts, and robustness testing. The method of this invention constructs a forward simulation path through triple fitting correction (meteorological data correction, power conversion correction, and scheduling standard evaluation); and utilizes the results of the third evaluation correction to form a feedback signal, dynamically adjusting the feature weighting and power calibration parameters in the forward path to form a closed-loop simulation; the digital twin engine integrates long-cycle factors such as component aging and a sudden event injection mechanism to simulate a high-fidelity dynamic environment. This invention enables dynamic evaluation of the adaptive learning capability of prediction schemes, robustness and resilience testing under extreme conditions, and provides a fair comparison environment and refined error attribution analysis, significantly improving the efficiency and depth of photovoltaic power prediction system research and development and verification.

Description

Technical Field

[0001] This invention relates to the fields of new energy power generation and digital simulation technology, and in particular to a photovoltaic power prediction simulation platform and simulation method based on fitting correction and digital twins. Background Technology

[0002] With the continuous increase in the proportion of renewable energy in the global energy structure, large-scale grid connection of new energy power generation, represented by photovoltaics, has become a trend. However, the intermittency and volatility of photovoltaic output pose serious challenges to the stable operation and economic dispatch of the power system. To improve grid friendliness and acceptance capacity, high-precision photovoltaic power forecasting technology has emerged. By accurately predicting the power generation of photovoltaic power plants over a future period, it provides key decision support for dispatching plans, reserve capacity allocation, and market transactions. Currently, mainstream forecasting technologies integrate mechanistic models based on meteorological physical processes with artificial intelligence models based on historical data mining, and verify and evaluate the algorithm performance through simulation platforms. These simulation platforms are typically based on historical meteorological and power data for backtesting and have become indispensable tools for the research and development and selection of forecasting algorithms.

[0003] However, existing photovoltaic power prediction simulation technologies still have significant shortcomings in simulating real-world operating environments and evaluating the dynamic performance of systems. First, traditional simulation platforms often employ open-loop, offline testing modes, meaning they input fixed historical datasets and output static prediction results. This mode cannot reproduce the closed-loop self-learning process of a real prediction system dynamically adjusting model parameters based on real-time error feedback during operation, leading to significant discrepancies between simulation evaluation results and actual online performance. Second, existing simulation methods typically treat the prediction system as a complete "black box," making it difficult to effectively decouple the total prediction error and clearly attribute it to factors such as weather forecast deviations, photovoltaic conversion model distortion, or changes in grid assessment rules, hindering targeted algorithm optimization. Furthermore, when comparing and selecting multiple solutions, the lack of standardized operating environment isolation mechanisms often leads to performance comparisons between different algorithms being influenced by non-technical factors such as computing resources and software dependency library versions, making it difficult to guarantee fairness. Finally, relying solely on historical data playback is insufficient to cover low-probability extreme events such as sudden cloud cover and equipment failure. Existing platforms generally lack an injection mechanism that integrates discrete disturbance events with continuous data streams, resulting in insufficient ability to assess the robustness and resilience of the prediction system under extreme conditions. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a photovoltaic power prediction simulation platform based on fitting correction and digital twins to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a photovoltaic power prediction simulation platform based on fitting correction and digital twins, comprising: The hybrid data source module is configured to access real historical data from photovoltaic power plants and generate virtual data based on the statistical characteristics of the real historical data. The scheme configuration and comparison module is configured to deploy at least one prediction scheme to be simulated and to compare the performance of the simulation results of each prediction scheme. The digital twin simulation engine module is connected to the hybrid data source module and the scheme configuration and comparison module. It includes a triple fitting correction logic unit, which is configured to execute the forward simulation path, generate the initial power prediction result and the third correction vector, and execute the feedback simulation path to update the simulation parameters using the third correction vector. The scenario script and robust testing module are configured to define simulation scenarios and inject discrete extreme events during the simulation process; The results comparison and visualization module is configured to display the simulation results of each prediction scheme.

[0007] As a preferred embodiment of the photovoltaic power prediction simulation platform for fitting correction and digital twins described in this invention, the scheme configuration and comparison module adopts containerization technology to provide an independent, resource-constrained operating environment for each prediction scheme and to fix the algorithm dependency library version to ensure the fairness of multi-scheme comparison.

[0008] As a preferred embodiment of the photovoltaic power prediction simulation platform for fitting correction and digital twins described in this invention, the digital twin simulation engine module operates using a dynamic clock-event dual-axis drive mechanism, including: The clock axis is used to control the simulation time to play back at normal speed or speedup and to drive a continuous data stream; The event axis is used to encode the scenario script and the extreme events generated by the robust testing module into a timestamp sequence through a message queue, and inject the timestamp sequence into the simulation process.

[0009] As a preferred embodiment of the photovoltaic power prediction simulation platform based on fitting correction and digital twins described in this invention, the digital twin simulation engine module further includes a long-period simulation unit for time-varying factors, which contains: The component aging register is used to accumulate the component performance degradation factor over time. The seasonal drift register is used to refresh the periodic code representing seasonal changes based on the simulation date; The attenuation coefficient and period encoding are used as parameters or features to participate in the operation of the triple fitting correction logic unit, simulating the long-term performance evolution of the photovoltaic power station.

[0010] Furthermore, this invention also provides a photovoltaic power prediction simulation method based on fitting correction and digital twins, the method comprising: S1: The input data for simulation is obtained through the hybrid data source module. The input data includes real historical data and virtual data generated based on the real historical data. S2: Deploy at least one photovoltaic power prediction scheme to be simulated through the scheme configuration and comparison module; S3: In the digital twin simulation engine, a closed-loop simulation process is executed for each prediction scheme. The closed-loop simulation process includes: S31: Execute the forward simulation path, sequentially perform the first fitting correction, the second fitting correction and the third evaluation correction on the input data, and generate the initial power prediction result and the third correction vector; S32: Execute the feedback simulation path, write the third correction vector back to the simulation unit parameters of the first fitting correction and the second fitting correction, and complete the closed-loop self-learning iteration; S4: Through the results comparison and visualization module, output and analyze the performance of each scheme in the closed-loop simulation process.

[0011] As a preferred embodiment of the photovoltaic power prediction simulation method based on fitting correction and digital twins described in this invention, the forward simulation path in S31 includes: The input meteorological data is first fitted and corrected to generate corrected usable meteorological data; Based on the available meteorological data, the physical model and the data-driven model are run in parallel and weighted and fused to perform a second fitting correction and generate an initial power prediction value. Based on the preset power grid dispatching standards, the initial power prediction value is evaluated and corrected in a third way, and the third correction vector is calculated.

[0012] As a preferred embodiment of the photovoltaic power prediction simulation method based on fitting correction and digital twins described in this invention, the execution feedback simulation path in S32 includes: The third correction vector is written back into the power calibration parameters used for the second fitting correction. A fine-tuning signal is calculated based on the third correction vector, and the fine-tuning signal is sent to the feature weighting parameters used for the first fitting correction.

[0013] As a preferred embodiment of the photovoltaic power prediction simulation method based on fitting correction and digital twins described in this invention, wherein: in step S3, the digital twin simulation engine supports online mode and accelerated playback mode, including: In online mode, the simulation clock is synchronized with the physical power plant clock, and the simulation time step is consistent with the data sampling period; In accelerated playback mode, the simulation clock advances at a preset rate greater than 1, continuously recalling historical data while maintaining the data timestamp order.

[0014] As a preferred embodiment of the photovoltaic power prediction simulation method based on fitting correction and digital twins described in this invention, wherein: in step S3, a step of superimposing simulations of module aging and seasonal drift is further included, the step comprising: Based on the simulation year, a fixed attenuation coefficient is added to the component aging parameters, and the attenuation coefficient is used as an offset in the calculation of the second fitting correction. Based on the current simulation date sequence, a periodic code is refreshed in the seasonal drift parameters, and the periodic code is introduced as an additional input feature into the calculation of the first fitting correction.

[0015] As a preferred embodiment of the photovoltaic power prediction simulation method based on fitting correction and digital twins described in this invention, in step S3, extreme events are injected in real time on the simulation timeline through a scenario script and robustness testing module; the extreme events include at least one of sudden cloud cover, sudden drop in irradiance, inverter tripping, module shading, or snow accumulation; when a meteorological event that changes irradiance or temperature is injected, the digital twin simulation engine synchronously calculates the theoretical power output under that condition and uses it as the measured value benchmark for the third evaluation correction.

[0016] Compared with existing technologies, the beneficial effects of the invention are: 1. This invention constructs a feedback closed-loop path from power prediction error (third evaluation correction) to meteorological data correction (first fitting correction) and power conversion correction (second fitting correction), which can simulate the process of online calibration and adaptive learning of the prediction scheme based on the power grid dispatch assessment results in real operation. This solves the technical gap of traditional open-loop testing methods, which can only evaluate static accuracy and cannot measure the dynamic adaptability of the model.

[0017] 2. The digital twin simulation engine of the present invention not only replays historical data, but also incorporates long-cycle time-varying factors such as photovoltaic module aging and seasonal efficiency drift into the simulation process through the built-in time-varying factor simulation unit; at the same time, through the dynamic clock-event dual-axis drive mechanism, it supports the precise injection of discrete disturbance events such as sudden weather events and equipment failures, so that the simulation environment can infinitely approximate the complex dynamic characteristics of the physical power station.

[0018] 3. Through scenario scripts and robust testing modules, this invention can customize and reproduce various extreme operating conditions (such as extreme weather and power grid disturbances), and observe the response, recovery process, and performance degradation of different prediction schemes under these shocks. This solves the problem that traditional testing methods are difficult to reproduce and quantify non-steady-state shock responses, providing a key testing tool for the development of high-reliability prediction systems.

[0019] 4. Furthermore, this invention employs containerization technology to provide a fair and isolated comparative testing environment for different prediction schemes, eliminating interference from differences in software and hardware environments. Simultaneously, the decoupled design of the triple fitting correction allows the platform to trace the final total prediction error back to whether it stems from meteorological data quality, model transformation accuracy, or external disturbances, providing clear guidance for iterative optimization of the algorithm. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a diagram of the photovoltaic power prediction simulation platform architecture based on fitting correction and digital twins according to an embodiment of the present invention; Figure 2 This is a flowchart of the triple fitting correction and feedback write-back process of the digital twin simulation engine according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the interactive interface for simulation scene configuration and extreme event injection according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a single-task simulation operation monitoring and error timing analysis interface according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the interface for multi-scheme parallel error attribution analysis and performance comparison according to an embodiment of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0025] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0026] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0027] Example 1 Reference Figure 1This is the first embodiment of the present invention, which provides a photovoltaic power prediction simulation platform for fitting correction and digital twins. The simulation platform is deployed on a high-performance server, with hardware configuration including an Intel Xeon Gold 6248R CPU, 128GB DDR4 memory, and an NVIDIA RTX A6000 GPU. The software environment is an Ubuntu 24.04 LTS operating system, containerized using Docker 20.10. Furthermore, it supports dynamic access to time-series databases (such as InfluxDB) and relational databases (such as MySQL). In this embodiment, a historical database, MySQL 8.0, is used as an example.

[0028] Specifically, 214 days of historical data from a photovoltaic power station were cleaned and stored in this historical database. The data table structure of this historical database is shown in Table 1.

[0029] Table 1 The data table above follows the following structure: <timestamp, station ID, parameter type, value>, to ensure seamless compatibility with SCADA, meteorological observation, and weather forecast data. A code example of this data table structure is shown below: CREATE TABLE `Simulation_measured_weather` ( `id` int NOT NULL AUTO_INCREMENT, `curtime` datetime NULL DEFAULT CURRENT_TIMESTAMP, `airpressure` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT 'Air pressure', `humidity` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT 'humidity', `radiation` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT 'radiation', `radiation_fz` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL, `scatter` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT 'scatter', `temperature` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT 'Temperature', `winddirection` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT 'wind direction', `windspeed` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT 'wind speed', `station_id` varchar(255) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL, PRIMARY KEY (`id`) USING BTREE ) ENGINE = InnoDB AUTO_INCREMENT = 75215 CHARACTER SET = utf8mb4COLLATE = utf8mb4_0900_ai_ci ROW_FORMAT = Dynamic; Furthermore, in the simulation platform of the present invention, the hybrid data source module is configured as the data throughput hub of the simulation platform, used to integrate real-world data and virtual construction data.

[0030] Furthermore, the hybrid data source module also includes a real data access unit, a virtual data generation unit, and a consistency verification unit.

[0031] Specifically, the real data access unit connects to the SCADA system data of the photovoltaic power station and the meteorological station data by connecting to the InfluxDB time-series database and the MySQL relational database.

[0032] Specifically, the virtual data generation unit generates virtual data to supplement extreme operating conditions based on the statistical characteristics (such as mean and standard deviation) of real historical data. For example, to simulate sudden changes in irradiance caused by cloud cover or strong winds, this unit incorporates an extreme weather disturbance model.

[0033] Furthermore, this extreme weather disturbance model is expressed as follows: Where k is the disturbance intensity coefficient; These are random numbers that follow a standard normal distribution, used to simulate scenarios such as cloud cover and strong winds. This indicates the irradiance under extreme conditions. This indicates the irradiance under historical conditions.

[0034] Specifically, the consistency verification unit dynamically compares the statistical characteristics of real and virtual data by setting a sliding window to ensure that the generated virtual scene conforms to the objective laws of the physical world.

[0035] Furthermore, the configuration and comparison module in the simulation platform of this invention supports parallel simulation of multiple schemes and quantifies the performance comparison of different schemes through a unified evaluation index system. Specifically, this module has the following functions and operations: Docker containers are used to isolate the runtime environments of different scenarios to ensure fairness. At the same time, each prediction scenario runs in an independent Docker container, sharing the host machine kernel, but with fully limited resources (CPU quota, memory limit, I / O bandwidth) to ensure a fair comparison with "same data, same resources, and same starting point".

[0036] Because the image layer solidification algorithm depends on the library version, it needs to automatically generate SHA-256 digests at startup to prevent runtime replacement.

[0037] The source code for calculating indicators, such as the latest version of the State Grid's "New Energy Power Prediction Assessment Method", is directly scripted to ensure version consistency at the source code level for indicator calculation.

[0038] Outputs visual charts (such as bar charts and error curves) and structured reports, supporting PDF / Excel export. Supports importing and exporting configuration files in JSON / YAML formats for easy batch solution management.

[0039] Furthermore, the digital twin simulation engine module in the simulation platform of the present invention is used to perform digital twin simulation on a specific object, namely, triple fitting correction prediction. Specifically, the digital twin simulation engine module includes a simulation control unit and a triple fitting correction logic unit.

[0040] Furthermore, the simulation control unit is mainly used to set up simulation scenarios and events.

[0041] Specifically, the simulation control unit adopts a dynamic clock-event dual-axis drive. That is, the clock axis supports two modes: "real timestamp playback every second" and "up to 1000× acceleration", which can be seamlessly switched; the event axis encodes discrete events such as cloud cover, inverter tripping, and snow accumulation into timestamp sequences and injects them into the simulation stream to achieve full-link synchronous operation of "meteorology-electricity-dispatch".

[0042] Furthermore, the simulation control unit also employs a long-period simulation method for time-varying factors, specifically including: Component aging: A fixed attenuation of -0.5% is added to the preset coefficient every year, which is superimposed on the online correction of the correction logic unit to truly reflect the coupling effect of "equipment aging + instantaneous self-adaptation"; Seasonal drift: An automatic refresh method with periodic coding is used to ensure that the model parameter trajectory from spring to summer to autumn to winter is traceable and auditable.

[0043] Furthermore, the triple fitting correction logic unit in the simulation platform of the present invention is used to fully reproduce the triple fitting correction logic of the photovoltaic power prediction system after the simulation scenario and events are set, including meteorological data correction, power conversion correction, scheduling evaluation correction and closed-loop feedback, supporting error decoupling and collaborative suppression.

[0044] In addition, the triple fitting correction logic unit also incorporates calculation logic consistent with the actual photovoltaic power prediction system.

[0045] Furthermore, the first fitting correction simulation unit in this triple fitting correction logic unit is used to simulate meteorological data deviation correction, and the specific calculation steps are as follows: Receive the actual raw data processed by the data acquisition and processing module of the photovoltaic power prediction system; Based on the types and parameters of the set scenarios and events, the virtual data simulator corrects the real original data by calculating factors such as the sky clearness factor, irradiance fluctuation rate, and temperature correction coefficient, and generates simulated virtual data.

[0046] In addition, the second fitting correction simulation unit can also generate available irradiance data for the component panel; available temperature data for the component panel; and fused available meteorological data.

[0047] Furthermore, the second fitting correction simulation unit in this triple fitting correction logic unit is used to simulate the power conversion process, which is as follows: The system receives the fused available meteorological data; generates the first type of power data; generates the second type of power data; generates the fused initial predicted power data; and processes the initial predicted power data based on the set scenarios and event types and parameters, such as photovoltaic module failure, inverter failure, combiner equipment failure, voltage events, scheduling instructions, planned maintenance, emergency handling, data quality events, etc., to generate simulated predicted power data.

[0048] Furthermore, the third fitting evaluation and correction simulation unit in this triple fitting correction logic unit is used to simulate scheduling evaluation and closed-loop feedback. The operation steps of this module are as follows: Evaluation criteria processing; Time window processing; Third correction vector operation; Furthermore, regarding the feedback update simulation unit in the simulation platform of this invention, this unit is used to implement the safe limiting and closed-loop write-back of the correction vector, and the specific operation steps are as follows: Correct vector limiting; Correct vector write-back; Furthermore, the control simulation process and evaluation standard module in the simulation platform of the present invention is used to support user-defined simulation step size, window length, evaluation indicators, etc.

[0049] Furthermore, the module includes a process orchestration engine and evaluation standard configurations.

[0050] Specifically, the process orchestration engine is mainly based on state machine design to support customizable processes such as "data loading → model configuration → simulation running → evaluation and comparison → result output".

[0051] Specifically, this evaluation standard is configured to support the import of power grid dispatch assessment standards and dynamically adjust the evaluation window length (e.g., ultra-short term 4 hours, short term 1-3 days, medium term 7-10 days).

[0052] Furthermore, the scenario script and robustness testing module in the simulation platform of the present invention is used to support the simulation of abnormal scenarios and events (such as cloud cover and equipment failure) and system resilience assessment, thereby verifying the stability of the platform under extreme conditions.

[0053] Furthermore, the scenario script and robust testing module also include a script editor and a test execution engine.

[0054] Specifically, the script editor is mainly responsible for providing a graphical interface that allows users to define and combine various abnormal events (such as sensor failures, extreme radiation fluctuations, and power rationing instructions) to form a complete test script.

[0055] Specifically, this test execution engine is used to drive the digital twin engine, inject events according to the script, and monitor the system's response and performance.

[0056] Furthermore, the result comparison and visualization module in the simulation platform of the present invention serves as the output and display window for the simulation platform results. It includes a multi-scheme dashboard, a quantitative evaluation report, and attribution analysis tools.

[0057] Specifically, a multi-scheme dashboard refers to displaying the prediction curves, error distributions, and correction vector trajectories of each scheme in parallel in the form of charts.

[0058] Specifically, the quantitative evaluation report is used to automatically generate and compare the performance indicators (RMSE, MAE, pass rate, etc.) of each solution, and supports export operations.

[0059] Specifically, attribution analysis tools, using visualization techniques, help users analyze the sources of error, thereby understanding the contribution of different correction modules.

[0060] Example 2 Reference Figure 2 This is a second embodiment of the present invention, which provides a photovoltaic power prediction simulation method based on fitting correction and digital twins. This method is executed on the simulation platform described in the present invention and includes: S1: Obtain input data for simulation through the hybrid data source module.

[0061] Specifically, this input data includes historical meteorological data (irradiance, temperature, wind speed, etc.) and power data of photovoltaic power plants read from a real database. Then, we generate virtual data using the platform's built-in virtual data simulator to simulate a "sudden cloud cover in a winter afternoon" scenario. A Gaussian perturbation is superimposed on the baseline irradiance sequence, calculated using the following formula: Where A is the shading strength; To obscure the center time; The duration of occlusion is controlled by the hour; in this embodiment, the value is set to 0.5. Let represent the simulated irradiance at time t. It is represented as the reference irradiance at time t.

[0062] S2: Deploy at least one photovoltaic power prediction scheme to be simulated through the scheme configuration and comparison module.

[0063] Furthermore, through the platform's graphical configuration interface, a preset Docker image is loaded, and three representative comparison schemes are configured. All comparison schemes are assigned the same number of CPU cores (e.g., 2 cores), memory limit (4GB), and I / O bandwidth to ensure the fairness of the evaluation environment. In this embodiment, the three comparison schemes are listed below: Option A (Benchmark Group - Traditional Physical Model): Only enable the physical model units in the second fitting correction module. Configuration parameters: Physical model weight w=1.0, disable data-driven model, disable closed-loop feedback path.

[0064] It should be noted that Scheme A is mainly used to simulate traditional open-loop physical modeling predictions.

[0065] Option B (Control Group - Offline Data-Driven): Enables both physical and data-driven models. Configuration parameters: Fusion weight w = 0.5, but disables the third evaluation correction and feedback update unit.

[0066] It should be noted that Option B is mainly used to simulate the current mainstream "physics + AI" hybrid model, but it lacks online self-learning capabilities.

[0067] Scheme C (Experimental Group - Closed-Loop Scheme of this Invention): Fully enable the "triple fitting correction" and "feedback update" functions of the digital twin engine. Configuration parameters: fusion weight w=0.5, self-learning rate λ=0.01, correction vector amplitude limit range is set to [0.85, 1.15].

[0068] It should be noted that Scheme C is used to verify the closed-loop evolution capability of the present invention.

[0069] Furthermore, the platform automatically starts a Docker container, limits the container to using 2 CPU cores and 4GB of memory, and loads fixed dependency libraries to complete the initialization of the simulation environment.

[0070] S3: In the digital twin simulation engine, a closed-loop simulation process is executed for each prediction scenario. The simulation engine starts a dual-axis drive, advances the simulation clock at a preset speedup (e.g., 100x), and executes the following sub-steps in a loop: S31: Execute the forward simulation path.

[0071] Specifically, the first fitting correction (meteorological correction): The simulation engine receives the raw meteorological data and processes it by combining the periodic coding features in the "Seasonal Drift Register". In this step, the clear sky factor, irradiance fluctuation rate, and temperature correction coefficient are calculated as the basis for correction, and the corrected usable meteorological data (usable irradiance and usable temperature) are output. The calculation formula refers to the GB / T 42766-2023 standard: Specifically, the sky clearness factor : in, This represents the total solar irradiance at the upper boundary of the atmosphere per hour. This refers to the hourly direct radiation dose to a horizontal surface. This represents the total solar irradiance per hour on a horizontal surface. This represents the total solar irradiance per hour on a horizontal surface.

[0072] Specifically, irradiance fluctuation rate : It should be noted that this irradiance fluctuation rate represents how much the solar energy changes within 15 minutes. On a sunny day, this change is small; however, if a cloud suddenly appears, the value will increase dramatically. Specifically, the temperature correction factor... : Where T represents the actual ambient temperature or the temperature of the module backsheet; This is the power temperature coefficient.

[0073] S31.2 Second Fit Correction (Power Conversion): Based on available meteorological data, the physical model and the data-driven model are run in parallel. In this step, the accumulated attenuation coefficient in the "Component Aging Register" (e.g., -0.5%×n in year n) is read and used as an offset in the physical model calculation to generate the initial power prediction value.

[0074] S31.3 Third Assessment Correction (Dispatch Assessment): The initial power prediction value is assessed based on the preset power grid dispatch standards (such as the latest assessment method). When a "power grid dispatch event" (such as an AGC power curtailment command) is detected, the system calculates the deviation between the theoretical power and the dispatch command and generates a third correction vector.

[0075] S32: Execute the feedback simulation path to complete the closed-loop self-learning iteration. Specifically, the simulation engine will generate a third correction vector. (After bandwidth limiting) Write back: Path 1: Generate the third correction vector The power calibration parameters are directly written back to the power calibration parameters of the second fitting correction unit to correct the power reference at the next time step.

[0076] Path 2: Calculate the fine-tuning signal The signal is then sent to the feature weighting parameter register of the first fitting correction unit to dynamically adjust the weights of the meteorological features.

[0077] S33: Real-time injection of extreme events Specifically, during the simulation, the scenario script module triggers events based on timestamps. For example, if an "inverter trip" event is injected at t=14:00, the digital twin engine immediately calculates the theoretical power output under the fault condition and uses it as a new measured value benchmark, forcing the prediction scheme to respond to the sudden disturbance through a closed-loop feedback path at the next moment.

[0078] S4. Through the result comparison and visualization module, output and analyze the performance of each scheme in the closed-loop simulation process.

[0079] Furthermore, after the simulation is complete, the platform generates a performance report. This report includes error timing analysis and attribution analysis.

[0080] Specifically, for error time series analysis, a comparison curve between predicted power and actual power is plotted to show the convergence speed of the prediction curve before and after the injection of extreme events (such as cloud cover).

[0081] Specifically, for attribution analysis, pie charts are used to illustrate the sources of error. For example, for closed-loop solution C, it demonstrates how it significantly reduces the proportion of error caused by "cloud influence" through a feedback mechanism, and quantifies it with metrics such as RMSE (root mean square error) and MAE (mean absolute error).

[0082] Example 3 Reference Figures 3 to 5 This is the third embodiment of the present invention, which provides a verification effect of the present invention's solution, specifically demonstrating the superiority of solution C (the closed-loop solution of the present invention) compared to solution A (traditional physical) and solution B (offline data driven).

[0083] Further reference Figure 3 Before starting the simulation, the test environment is defined through the "Scene Parameter Configuration" interface to verify the "dual-axis drive mechanism" in the present invention: Clock axis configuration: In the "Time-driven configuration" panel, set the simulation time step to 60 seconds and enable "100x" dynamic speed to achieve fast playback of long-cycle data.

[0084] Environmental parameter initialization: In the "Environmental Parameters" column, set the reference irradiance to 1000 W / m² and the reference temperature to 25.00°C to establish a standard test reference.

[0085] Event Injection: In the "Event Injection Configuration" panel, configure a "Grid Dispatch Event" to test the system's responsiveness to grid commands. Select "AGC Dispatch" as the event subclass, set the target power limit to 50.00 MW, the ramp rate limit to 2.00 MW / min, and the response delay to 5 seconds. Click "Add Current Event" to encode and inject it into the simulation stream.

[0086] It should be noted that the aforementioned "dual-axis drive mechanism" activates the third fitting evaluation and correction unit in the digital twin engine, forcing the prediction system to respond to the AGC power-limiting command during simulation; otherwise, significant evaluation errors will occur. Furthermore, refer to... Figure 4 Once a single simulation is complete, the "File Selection and Analysis" interface will appear. Task queue monitoring: The "Simulation Task List" displays the currently executing Task IDs (such as T001, T002) in real time, covering various weather types such as "sunny forecast", "cloudy forecast" and "seasonal forecast", proving that the platform has the ability to handle multiple scenarios concurrently.

[0087] Curve fitting observation: In the "Power Prediction Curve" view, the blue curve closely matches the red curve. Especially in the irradiance fluctuation range of 11.5 hours to 17.3 hours, the curve did not show significant overshoot, indicating that the closed-loop feedback mechanism effectively suppressed model drift.

[0088] Error time series tracking: The "Error Analysis View" shows that the prediction error was controlled within ±1.5% for most of the time period, which verifies the stability of the model in the dynamic process.

[0089] Furthermore, in the "Attribution Analysis Configuration" interface, the simulation platform will automatically perform parallel comparisons of comparison schemes A, B, and C.

[0090] Error attribution decoupling (pie chart analysis): The simulation platform uses intermediate variables corrected by triple fitting to decouple the total prediction error into four dimensions, which are then displayed as a pie chart: Option 1 (left column): It shows that "Cloud Impact" accounts for as much as 35%, and "Model Error" accounts for 20%. This indicates that the pure physical model cannot effectively cope with sudden changes in cloud cover.

[0091] Option 2 (middle column): Although a data-driven model has been introduced, "Cloud Impact" still accounts for 40%, indicating that the offline model is still not adaptable enough to sudden weather changes.

[0092] Option 3 (right column): Through a closed-loop feedback mechanism, the "Model Error" is reduced to 15%, and the "Data Quality" error is effectively identified.

[0093] It should be noted that by writing back the correction vector, we can make the overall error distribution of the prediction system more uniform.

[0094] Specifically, in the "Error Distribution" and "Main Sources of Error" text fields below, the platform will automatically generate the conclusion: After introducing the feedback mechanism, Scheme 3 effectively suppressed the nonlinear error caused by meteorological fluctuations.

[0095] It should be noted that, through comparison Figure 5 The error attribution analysis shown clearly demonstrates that Scheme C (the closed-loop scheme of this invention) not only has the lowest total error (RMSE), but also the proportion of errors caused by the model itself and cloud layer influence is significantly lower than that of Schemes A and B. This proves the significant advantages of the simulation platform and simulation method of this invention in simulating closed-loop self-learning, decoupling errors, and evaluating system robustness.

[0096] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0097] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0098] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0099] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0100] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0101] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A photovoltaic power prediction simulation platform based on fitting correction and digital twins, characterized in that, include: The hybrid data source module is configured to access real historical data from photovoltaic power plants and generate virtual data based on the statistical characteristics of the real historical data. The scheme configuration and comparison module is configured to deploy at least one prediction scheme to be simulated and to compare the performance of the simulation results of each prediction scheme. The digital twin simulation engine module is connected to the hybrid data source module and the scheme configuration and comparison module. It includes a triple fitting correction logic unit, which is configured to execute the forward simulation path, generate the initial power prediction result and the third correction vector, and execute the feedback simulation path to update the simulation parameters using the third correction vector. The scenario script and robust testing module are configured to define simulation scenarios and inject discrete extreme events during the simulation process; The results comparison and visualization module is configured to display the simulation results of each prediction scheme.

2. The photovoltaic power prediction simulation platform based on fitting correction and digital twin as described in claim 1, characterized in that, The scheme configuration and comparison module adopts containerization technology to provide an independent, resource-constrained operating environment for each prediction scheme and fixes the algorithm dependency library version to ensure the fairness of multi-scheme comparison.

3. The photovoltaic power prediction simulation platform based on fitting correction and digital twin as described in claim 1, characterized in that, The digital twin simulation engine module operates using a dynamic clock-event dual-axis drive mechanism, including: The clock axis is used to control the simulation time to play back at normal speed or speedup and to drive a continuous data stream; The event axis is used to encode the scenario script and the extreme events generated by the robust testing module into a timestamp sequence through a message queue, and inject the timestamp sequence into the simulation process.

4. The photovoltaic power prediction simulation platform based on fitting correction and digital twin as described in claim 1, characterized in that, The digital twin simulation engine module further includes a long-period simulation unit for time-varying factors, which contains: The component aging register is used to accumulate the component performance degradation factor over time. The seasonal drift register is used to refresh the periodic code representing seasonal changes based on the simulation date; The attenuation coefficient and period encoding are used as parameters or features to participate in the operation of the triple fitting correction logic unit, simulating the long-term performance evolution of the photovoltaic power station.

5. A photovoltaic power prediction simulation method based on fitting correction and digital twins, characterized in that, include: S1: The input data for simulation is obtained through the hybrid data source module. The input data includes real historical data and virtual data generated based on the real historical data. S2: Deploy at least one photovoltaic power prediction scheme to be simulated through the scheme configuration and comparison module; S3: In the digital twin simulation engine, a closed-loop simulation process is executed for each prediction scheme. The closed-loop simulation process includes: S31: Execute the forward simulation path, sequentially perform the first fitting correction, the second fitting correction and the third evaluation correction on the input data, and generate the initial power prediction result and the third correction vector; S32: Execute the feedback simulation path, write the third correction vector back to the simulation unit parameters of the first fitting correction and the second fitting correction, and complete the closed-loop self-learning iteration; S4: Through the results comparison and visualization module, output and analyze the performance of each scheme in the closed-loop simulation process.

6. The photovoltaic power prediction simulation method based on fitting correction and digital twin as described in claim 5, characterized in that, The forward simulation path in S31 includes: The input meteorological data is first fitted and corrected to generate corrected usable meteorological data; Based on the available meteorological data, the physical model and the data-driven model are run in parallel and weighted and fused to perform a second fitting correction and generate an initial power prediction value. Based on the preset power grid dispatching standards, the initial power prediction value is evaluated and corrected in a third way, and the third correction vector is calculated.

7. The photovoltaic power prediction simulation method based on fitting correction and digital twin as described in claim 6, characterized in that, The execution feedback simulation path in S32 includes: The third correction vector is written back into the power calibration parameters used for the second fitting correction. A fine-tuning signal is calculated based on the third correction vector, and the fine-tuning signal is sent to the feature weighting parameters used for the first fitting correction.

8. The photovoltaic power prediction simulation method based on fitting correction and digital twin as described in claim 5, characterized in that, In S3, the digital twin simulation engine supports online mode and accelerated playback mode, including: In online mode, the simulation clock is synchronized with the physical power plant clock, and the simulation time step is consistent with the data sampling period; In accelerated playback mode, the simulation clock advances at a preset rate greater than 1, continuously recalling historical data while maintaining the data timestamp order.

9. The photovoltaic power prediction simulation method based on fitting correction and digital twin as described in claim 5, characterized in that, S3 further includes a step of superimposed simulation of component aging and seasonal drift, the step comprising: Based on the simulation year, a fixed attenuation coefficient is added to the component aging parameters, and the attenuation coefficient is used as an offset in the calculation of the second fitting correction. Based on the current simulation date sequence, a periodic code is refreshed in the seasonal drift parameters, and the periodic code is introduced as an additional input feature into the calculation of the first fitting correction.

10. The photovoltaic power prediction simulation method based on fitting correction and digital twin as described in claim 5, characterized in that, In S3, extreme events are injected in real time on the simulation timeline through the scenario script and robust testing module; the extreme events include at least one of sudden cloud cover, sudden drop in irradiance, inverter tripping, component shading or snow accumulation; when a meteorological event that changes irradiance or temperature is injected, the digital twin simulation engine synchronously calculates the theoretical power output under the condition and uses it as the measured value benchmark for the third evaluation correction.

Images