Distributed photovoltaic power uncertainty quantification and prediction method based on multi-modal fusion and adaptive learning

Abstract

The application discloses a distributed photovoltaic power uncertainty quantification and prediction method based on multi-modal fusion and adaptive learning, comprising the following steps: S1, acquiring first snow cover data; S2, collecting second snow influence data; S3, completing multi-modal data fusion in a three-dimensional feature space; S4, dividing an anomaly detection domain according to a preset threshold condition in the three-dimensional feature space; S5, for the labeled area in the anomaly detection domain, iteratively generating a power prediction band based on Monte Carlo simulation; S6, activating a dynamic compensation algorithm, calculating a compensation coefficient and feeding the coefficient back to the power prediction band simulation input; S7, forming an anomaly detection-compensation adjustment-detection domain adaptive closed-loop iteration mechanism. The application quantifies the on-site uncertainty such as snow and hot spots in photovoltaic power prediction, and significantly improves the prediction accuracy and system response speed through the dynamic compensation and threshold adaptive mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning. Background Technology

[0002] Currently, distributed photovoltaic (PV) power plants face significant challenges in power generation and prediction deviations under winter snow cover conditions. Snow not only obstructs PV panels, reducing effective irradiance, but also induces localized hot spots by altering the surface thermal conductivity of the modules, accelerating module aging or damage. Existing technologies typically rely on a single data source (such as weather station data or satellite imagery) for power prediction, lacking the fine-grained capture of microscopic local thermal anomalies. This results in a delayed response of prediction models to unusual scenarios (such as sudden snowfall), making it difficult to compensate for errors in a timely manner. Furthermore, traditional prediction methods are mostly based on static models, unable to dynamically adjust the prediction bandwidth according to real-time environmental changes, making them prone to abnormal power fluctuations that cannot be reflected in the output power prediction in a timely manner.

[0003] In thermodynamic modeling, current methods mostly ignore the influence of physical mechanisms such as snow crystal structure and phase transition processes on the thermodynamic state of photovoltaic panels. Regarding data fusion, single-scale (macroscopic or microscopic) data is used in isolation, failing to balance breadth of coverage with local accuracy. In terms of algorithm adaptability, existing uncertainty quantification methods mostly use fixed parameters, lacking the ability to self-correct in real time based on drastic environmental changes (such as snow melting rates). These shortcomings result in photovoltaic power plants experiencing problems during snow cover periods, including excessively wide confidence intervals for power prediction, lagging dynamic compensation mechanisms, and insufficient response to local thermal anomalies. 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 problems existing in the above-mentioned methods for quantitative prediction of distributed photovoltaic power uncertainty based on multimodal fusion and adaptive learning, this invention is proposed.

[0006] Therefore, the purpose of this invention is to provide a distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning, which quantifies the on-site uncertainties such as snow accumulation and hot spots in photovoltaic power prediction, and significantly improves prediction accuracy and system response speed through dynamic compensation and threshold adaptive mechanism.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning, comprising the following steps:

[0008] S1. Obtain first snow cover data. The first snow cover data is collected by a satellite remote sensing module and includes shortwave infrared band reflectivity information and corresponding geographic spatial coordinates.

[0009] S2. Collect the second snow impact data, which is collected through an edge computing node array;

[0010] S3. Perform multimodal data fusion in a three-dimensional feature space, where the X-axis of the three-dimensional feature space is snow thickness, the Y-axis is surface temperature, and the Z-axis is effective irradiance. The mapping and association between the first snow cover data and the second snow impact data are realized through interpolation and registration.

[0011] S4. In the three-dimensional feature space, an anomaly detection domain is divided according to a preset threshold condition, and data falling into the detection domain is automatically marked with compensation requirements.

[0012] S5. For the marked regions in the anomaly detection domain, a power prediction band is generated iteratively based on Monte Carlo simulation to quantify the prediction uncertainty;

[0013] S6. In real-time monitoring, when the heat flux rate of the phase change material is detected to exceed the preset threshold, the dynamic compensation algorithm is activated, the compensation coefficient is calculated and the coefficient is fed back to the power prediction band analog input to correct the prediction result in real time.

[0014] S7. The boundary conditions of the anomaly detection domain are dynamically updated based on the predicted bias and the actual power output bias after compensation, forming an adaptive closed-loop iterative mechanism of anomaly detection, compensation adjustment and detection domain.

[0015] As a preferred embodiment of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning described in this invention, in step S1, when the integrity of the first snow cover data is lower than a preset quality threshold, it automatically switches to using only the second snow cover impact data; otherwise, it fuses the two data as described in step S3.

[0016] As a preferred embodiment of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning described in this invention, the following step is added after step S2: when the surface temperature change rate in the second snow impact data exceeds a preset rate threshold, the sampling frequency of the edge computing node is automatically increased; otherwise, the original sampling frequency is maintained.

[0017] As a preferred embodiment of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning described in this invention, in step S4, if the number of marked abnormal detection domains continuously exceeds a preset number N, the threshold condition is successively reduced to expand the detection domain, and then step S5 is entered; otherwise, the original threshold is maintained and step S5 is continued.

[0018] As a preferred embodiment of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning described in this invention, the threshold conditions include a temperature threshold and a thickness threshold. When the temperature threshold trigger frequency exceeds the threshold trigger frequency in step S4, the sensitivity of the thickness threshold judgment is further improved to optimize the anomaly detection range.

[0019] As a preferred embodiment of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning described in this invention, in step S7, if the deviation between the predicted bias and the actual output bias after compensation shows a continuous decreasing trend, the boundary of the anomaly detection domain is tightened; otherwise, the boundary of the anomaly detection domain is widened by a preset ratio; when the cumulative widening ratio of the detection domain boundary exceeds the set upper limit after executing step S7 for M consecutive times, model retraining is triggered to recalibrate the three-dimensional feature space mapping relationship.

[0020] As a preferred embodiment of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning described in this invention, the dynamic compensation algorithm in step S6 includes adaptive updating of the recursive compensation coefficient based on historical prediction errors to ensure that the compensation coefficient converges in real time as the environment changes.

[0021] A distributed photovoltaic power uncertainty quantification prediction system based on multimodal fusion and adaptive learning, characterized in that it includes:

[0022] Satellite remote sensing module: acquires shortwave infrared reflectance data and calculates snow cover thickness;

[0023] Edge computing node array: Installed on the surface of each photovoltaic module, it collects data in real time from temperature sensors, local irradiance sensors and hot spot detection cameras;

[0024] 3D Feature Space Construction Module: Constructs a data fusion space using snow thickness on the X-axis, component surface temperature on the Y-axis, and effective irradiance on the Z-axis;

[0025] Interpolation and registration module: Performs spatial registration and interpolation on macroscopic and microscopic data to achieve mapping and association of data at different resolutions;

[0026] Anomaly detection domain segmentation module: Based on preset temperature thresholds and thickness thresholds, it automatically identifies power anomaly areas in the feature space and marks the areas that need compensation;

[0027] Monte Carlo simulation module: Runs N random sampling simulations on the marked region to generate a power prediction band;

[0028] Phase change material heat flow monitoring module: Monitors the heat flow rate of the component through phase change material patch and heat flow rate sensor, and triggers dynamic compensation when it exceeds a preset threshold;

[0029] Dynamic compensation algorithm module: recursively updates the compensation coefficients based on historical prediction errors and corrects the input simulation in real time;

[0030] Closed-loop iteration and threshold adaptation module: Based on the deviation between the compensated predicted band and the actual output, dynamically tighten or loosen the detection domain boundary, and trigger model retraining when the loosening exceeds the limit multiple times.

[0031] The online optimization module for reinforcement learning updates model parameters in real time based on compensation feedback and state-action-reward design, thereby improving prediction and compensation performance.

[0032] The beneficial effects of this invention are:

[0033] This invention combines satellite remote sensing and multimodal data from edge computing nodes to effectively capture changes in the surface of photovoltaic modules and environmental snow conditions, thereby improving the adaptability of the prediction model to extreme operating conditions.

[0034] By fusing three-dimensional feature space, the effects of snow thickness, temperature and irradiance on power are accurately quantified, and a reliable power prediction band is obtained using the Monte Carlo method.

[0035] A dynamic compensation algorithm based on phase change material heat flow detection was used to achieve real-time correction of the impact of sudden hot spots or snow accumulation on site.

[0036] The adaptive threshold and closed-loop iteration mechanism enable the system to automatically adjust the detection domain based on error feedback, thereby improving the overall prediction accuracy.

[0037] The introduction of trigger retraining and reinforcement learning strategies ensures continuous model optimization and online adaptive capabilities, meeting the operational requirements of large-scale distributed photovoltaic power plants. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the 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:

[0039] Figure 1 This is a schematic diagram illustrating the steps of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning according to the present invention. Detailed Implementation

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] 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.

[0042] 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.

[0043] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0044] Reference Figure 1 This paper presents a method for quantitative prediction of distributed photovoltaic power uncertainty based on multimodal fusion and adaptive learning, including the following steps:

[0045] S1. Obtain the first snow cover data. The first snow cover data is collected through a satellite remote sensing module and includes shortwave infrared band reflectivity information and corresponding geospatial coordinates.

[0046] S2. Collect the second snow impact data. The second snow impact data is collected through an edge computing node array.

[0047] S3. Multimodal data fusion is completed in a three-dimensional feature space. The X-axis of the three-dimensional feature space is snow thickness, the Y-axis is surface temperature, and the Z-axis is effective irradiance. The mapping and association between the first snow cover data and the second snow impact data are realized through interpolation and registration.

[0048] S4. In the three-dimensional feature space, anomaly detection domains are divided according to preset threshold conditions, and data falling into the detection domains are automatically marked with compensation requirements.

[0049] S5. For the marked regions in the anomaly detection domain, power prediction bands are generated iteratively based on Monte Carlo simulation to quantify prediction uncertainty;

[0050] S6. In real-time monitoring, when the heat flux rate of the phase change material is detected to exceed the preset threshold, the dynamic compensation algorithm is activated, the compensation coefficient is calculated and the coefficient is fed back to the power prediction band analog input to correct the prediction result in real time.

[0051] S7. The boundary conditions of the anomaly detection domain are dynamically updated based on the predicted bias and the actual power output bias after compensation, forming an adaptive closed-loop iterative mechanism of anomaly detection, compensation adjustment and detection domain.

[0052] Step S1: Obtaining the first snow cover data includes:

[0053] Equipment and modules: Ground reflectivity data are acquired using a shortwave infrared (SWIR) band imager via a low-Earth orbit / medium-Earth orbit meteorological satellite positioned above or near a photovoltaic power station.

[0054] Data content: The reflectance of each pixel in the 1.5–2.5μm band and the corresponding geospatial coordinates (longitude, latitude, and elevation) within the acquisition area are collected and then subjected to atmospheric and geometric corrections to generate an array of "initial snow cover thickness values".

[0055] Preprocessing: Time registration is performed on data from multiple satellites at the same time, and reflectivity is mapped to snow thickness estimates using an empirical model, with a resolution of up to 30m×30m.

[0056] Step S2: Collecting the second snow impact data includes:

[0057] Edge computing node array: Edge nodes equipped with temperature sensors (thermocouples or infrared arrays), irradiance sensors and GPS coordinate acquisition modules are deployed behind or near each group of photovoltaic modules. Each node has local short-term storage and preliminary computing capabilities.

[0058] Sampling content:

[0059] Surface temperature distribution of photovoltaic modules: obtained by infrared imaging (640×480 resolution) or dot matrix temperature sensing;

[0060] Local irradiance: The actual irradiance on each component was measured using a high-sensitivity photodiode array;

[0061] Hot spot location coordinates: Extract the hot spot region from the temperature distribution image by threshold segmentation and record its center coordinates.

[0062] Quality control: The node performs time-series filtering on temperature data and performs occlusion removal (abrupt point removal) on irradiance.

[0063] Specifically, in step S1, when the integrity of the first snow cover data is lower than the preset quality threshold, it automatically switches to using only the second snow cover impact data; otherwise, it merges the two data according to step S3; after step S2, the following step is added: when the surface temperature change rate in the second snow cover impact data exceeds the preset rate threshold, it automatically increases the sampling frequency of the edge computing node; otherwise, it maintains the original sampling frequency.

[0064] Furthermore, step S3: fusion of multimodal data in the three-dimensional feature space includes:

[0065] Define a coordinate system: Construct a three-dimensional feature space, with the X-axis representing the snow thickness. The Y-axis represents the surface temperature of the component. The Z-axis represents the effective irradiance. ;

[0066] Registration and mapping:

[0067] For each edge node coordinate in the second snow impact data, find its corresponding reflectance value in the satellite pixel;

[0068] Satellite pixel snow thickness values ​​are mapped to edge node space using bilinear / cubic spline interpolation;

[0069] Establish mapping function The fused density function is constructed by interpolation in the three-dimensional feature space.

[0070] In step S4, if the number of marked abnormal detection domains exceeds the preset number N consecutively, the threshold condition is lowered sequentially to expand the detection domain before proceeding to step S5; otherwise, the original threshold is maintained and step S5 is executed. Specifically, the threshold condition includes a temperature threshold and a thickness threshold. When the temperature threshold trigger frequency exceeds the threshold trigger frequency in step S4, the sensitivity of the thickness threshold judgment is further improved to optimize the range of abnormal detection domains.

[0071] Specifically, step S4: anomaly detection domain division and labeling includes:

[0072] Threshold conditions: Set snow thickness thresholds in three-dimensional space. Temperature threshold Irradiance threshold ;

[0073] Detection domain: Data points that meet any one or a combination of conditions fall into the anomaly detection domain. ;

[0074] Labeling: The system automatically labels the interior points with "compensation requirements" and generates a corresponding compensation priority sequence;

[0075] If the number of outliers detected in N consecutive iterations exceeds a preset threshold... Then decrease sequentially. and To expand ;

[0076] When the temperature threshold trigger frequency exceeds a certain number of times, increase the sensitivity of the snow thickness threshold (i.e., reduce the sensitivity). (Step size) to optimize the detection domain.

[0077] Step S5: Monte Carlo simulation quantization prediction band generation includes:

[0078] Initial distribution assumption: Based on historical observation errors at the same location and under the same meteorological conditions, a probability distribution of power prediction errors is constructed. ;

[0079] Simulation procedure: Randomly select 10,000–100,000 groups within the labeled domain. Substitute the sample into the photovoltaic module power output model:

[0080]

[0081] in For nominal efficiency, , For snow thickness and temperature attenuation coefficient, Reference temperature;

[0082] Prediction band: The upper and lower percentiles of the power sample (e.g., 5%–95%) are taken as the prediction band interval to achieve uncertainty quantification.

[0083] Step S6: Dynamic compensation algorithm and real-time correction includes:

[0084] Phase change material heat flux monitoring: A phase change material (PCM) sensing layer is placed behind the component or at the junction box to monitor the heat flux rate in real time. ;

[0085] Threshold trigger: when When this occurs, the dynamic compensation algorithm is triggered to calculate the compensation coefficient:

[0086]

[0087] in: This is the initial value of the compensation coefficient used in the k-th iteration; The compensation coefficient is calculated in the (k+1)th iteration;

[0088] For learning rate, For the actual output, The median power of the current forecast band; It is the actual power output at time t, obtained from data monitored by field sensors or inverters; It is the predicted power value at time t, taken from the median (or other statistical indicator, such as the mean) of the current prediction band.

[0089] After each iteration, based on the historical error sequence Recursive update With the initial This ensures that the compensation coefficient converges rapidly to environmental changes.

[0090] Furthermore, in step S7, if the deviation between the predicted band after compensation and the actual output shows a continuous decreasing trend, the boundary of the anomaly detection domain is tightened; otherwise, the boundary of the anomaly detection domain is widened by a preset ratio; when the cumulative widening ratio of the detection domain boundary exceeds the set upper limit after executing step S7 for M consecutive times, model retraining is triggered to recalibrate the three-dimensional feature space mapping relationship.

[0091] Specifically, step S7: closed-loop self-iterating includes:

[0092] Boundary update: Based on the error in the current prediction after compensation. Deviation from actual Trend of change:

[0093] If both show a continuous downward trend, then tighten the detection domain boundary and increase... , ;

[0094] Otherwise, use the preset ratio. Relax the boundaries to ensure full coverage of anomalies.

[0095] Adaptive closed loop: The above detection-compensation-boundary update cycle continues until the prediction accuracy meets the set standard or the maximum number of iterations is reached.

[0096] Specific scenario examples:

[0097] A large-scale photovoltaic power station in a mountainous area:

[0098] 1. System Deployment and Parameter Settings

[0099] Satellite remote sensing module:

[0100] The Sentinel-2 satellite SWIR band (1.6) was selected. Data, spatial resolution 20m;

[0101] Atmospheric correction was performed using the Sen2Cor tool to generate an array of initial snow thickness values;

[0102] Empirical model: Snow thickness ,in , .

[0103] Edge computing nodes:

[0104] Fifty nodes are deployed, evenly distributed behind a 10MW string module cluster;

[0105] Temperature sensing: FLIR infrared array, resolution 640×480, temperature accuracy ±0.5℃;

[0106] Irradiance sensing: photodiode array, range 0–1500 W / m 2 Accuracy ±5W / m 2 ;

[0107] Phase change material layer sensing: heat flux threshold 200W / m 2 .

[0108] 2. Multimodal fusion and threshold conditions

[0109] Three-dimensional feature space coordinates:

[0110] Snow thickness ;

[0111] Surface temperature ;

[0112] Effective irradiance W / m 2

[0113] Initial threshold setting:

[0114] , , W / m 2 ;

[0115] 3. Anomaly Detection and Monte Carlo Simulation

[0116] Detection domain division:

[0117] When a node measures , , W / m 2 ,fall into Automatically mark compensation requirements;

[0118] If the number of detected domains exceeds 10% for three consecutive days, the trigger threshold will be lowered. , .

[0119] Monte Carlo simulation:

[0120] A sample size of 50,000 groups was selected, and the power model parameters were... , =0.004 / cm =0.005 / ;

[0121] Generate a prediction band of 5%-95%, for example, if the median prediction for a certain iteration is 8.2MW, the range is... .

[0122] 4. Dynamic compensation and closed-loop iteration

[0123] Compensation Triggered:

[0124] Heat flow rate was detected at a certain moment Calculate the initial compensation coefficient ;

[0125] When the actual output is 8.1MW, the predicted median is 8.2MW, and the error is -0.1MW, update.

[0126]

[0127] Boundary update:

[0128] After 5 iterations, the prediction error decreased from Down to The detection domain is tightened to , ;

[0129] Cumulative relaxation ratio This triggers model retraining;

[0130] 5. Performance

[0131] Within 30 days, this method reduced the daily root mean square error (RMSE) of forecasts from 6.5% to 3.1%.

[0132] For extreme snow accumulation (snow depth > 10 cm), the peak error is reduced from 12% to 5%.

[0133] This invention also includes a distributed photovoltaic power uncertainty quantification prediction system based on multimodal fusion and adaptive learning, comprising:

[0134] Satellite remote sensing module: acquires shortwave infrared reflectance data and calculates snow cover thickness;

[0135] Edge computing node array: Installed on the surface of each photovoltaic module, it collects data in real time from temperature sensors, local irradiance sensors and hot spot detection cameras;

[0136] 3D Feature Space Construction Module: Constructs a data fusion space using snow thickness on the X-axis, component surface temperature on the Y-axis, and effective irradiance on the Z-axis;

[0137] Interpolation and registration module: Performs spatial registration and interpolation on macroscopic and microscopic data to achieve mapping and association of data at different resolutions;

[0138] Anomaly detection domain segmentation module: Based on preset temperature thresholds and thickness thresholds, it automatically identifies power anomaly areas in the feature space and marks the areas that need compensation;

[0139] Monte Carlo simulation module: Runs N random sampling simulations on the marked region to generate a power prediction band;

[0140] Phase change material heat flow monitoring module: Monitors the heat flow rate of the component through phase change material patch and heat flow rate sensor, and triggers dynamic compensation when it exceeds a preset threshold;

[0141] Dynamic compensation algorithm module: recursively updates the compensation coefficients based on historical prediction errors and corrects the input simulation in real time;

[0142] Closed-loop iteration and threshold adaptation module: Based on the deviation between the compensated predicted band and the actual output, dynamically tighten or loosen the detection domain boundary, and trigger model retraining when the loosening exceeds the limit multiple times.

[0143] The online optimization module for reinforcement learning updates model parameters in real time based on compensation feedback and state-action-reward design, thereby improving prediction and compensation performance.

[0144] The present invention also provides a computer device applicable to a distributed photovoltaic power uncertainty quantification prediction system based on multimodal fusion and adaptive learning, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the distributed photovoltaic power uncertainty quantification prediction system based on multimodal fusion and adaptive learning as proposed in the above embodiments.

[0145] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0146] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements a distributed photovoltaic power uncertainty quantification prediction system based on multimodal fusion and adaptive learning, as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for quantitative prediction of distributed photovoltaic power uncertainty based on multimodal fusion and adaptive learning, characterized in that, Includes the following steps: S1. Obtain first snow cover data. The first snow cover data is collected by a satellite remote sensing module and includes shortwave infrared band reflectivity information and corresponding geographic spatial coordinates. S2. Collect the second snow impact data, which is collected through an edge computing node array; S3. Perform multimodal data fusion in a three-dimensional feature space, where the X-axis of the three-dimensional feature space is snow thickness, the Y-axis is surface temperature, and the Z-axis is effective irradiance. The mapping and association between the first snow cover data and the second snow impact data are realized through interpolation and registration. S4. In the three-dimensional feature space, an anomaly detection domain is divided according to a preset threshold condition, and data falling into the detection domain is automatically marked with compensation requirements. S5. For the marked regions in the anomaly detection domain, a power prediction band is generated iteratively based on Monte Carlo simulation to quantify the prediction uncertainty; S6. In real-time monitoring, when the heat flux rate of the phase change material is detected to exceed the preset threshold, the dynamic compensation algorithm is activated, the compensation coefficient is calculated and the coefficient is fed back to the power prediction band analog input to correct the prediction result in real time. S7. Dynamically update the boundary conditions of the anomaly detection domain based on the predicted band deviation after compensation and the actual power output deviation, forming an anomaly detection-compensation adjustment-detection domain adaptive closed-loop iterative mechanism. In step S1, when the integrity of the first snow cover data is lower than a preset quality threshold, the system automatically switches to using only the second snow cover impact data; otherwise, the two data are merged as described in step S3. After step S2, the following step is added: when the surface temperature change rate in the second snow impact data exceeds a preset rate threshold, the sampling frequency of the edge computing node is automatically increased; otherwise, the original sampling frequency is maintained.

2. The distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning according to claim 1, characterized in that: In step S4, if the number of marked abnormal detection domains exceeds the preset number N consecutively, the threshold condition is lowered sequentially to expand the detection domain, and then the process proceeds to step S5. Otherwise, maintain the original threshold and continue with step S5.

3. The distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning according to claim 2, characterized in that: The threshold conditions include a temperature threshold and a thickness threshold. When the temperature threshold trigger frequency exceeds the threshold trigger frequency in step S4, the sensitivity of the thickness threshold judgment is further improved to optimize the anomaly detection range.

4. The distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning according to claim 1, characterized in that: In step S7, if the deviation between the predicted band after compensation and the actual output deviation shows a continuous decreasing trend, the boundary of the anomaly detection domain is tightened; otherwise, the boundary of the anomaly detection domain is widened by a preset ratio. When the cumulative widening ratio of the detection domain boundary exceeds the set upper limit after executing step S7 for M consecutive times, model retraining is triggered to recalibrate the three-dimensional feature space mapping relationship.

5. The distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning according to claim 1, characterized in that: The dynamic compensation algorithm in step S6 includes adaptive updating of the recursive compensation coefficient based on historical prediction errors to ensure that the compensation coefficient converges in real time as the environment changes.

6. A distributed photovoltaic power uncertainty quantification prediction system based on multimodal fusion and adaptive learning, characterized in that, include: Satellite remote sensing module: acquires shortwave infrared reflectance data and calculates snow cover thickness; Edge computing node array: Installed on the surface of each photovoltaic module, it collects data in real time from temperature sensors, local irradiance sensors and hot spot detection cameras; 3D Feature Space Construction Module: Constructs a data fusion space using snow thickness on the X-axis, component surface temperature on the Y-axis, and effective irradiance on the Z-axis; Interpolation and registration module: Performs spatial registration and interpolation on macroscopic and microscopic data to achieve mapping and association of data at different resolutions; Anomaly detection domain segmentation module: Based on preset temperature thresholds and thickness thresholds, it automatically identifies power anomaly areas in the feature space and marks the areas that need compensation; Monte Carlo simulation module: Runs N random sampling simulations on the marked region to generate a power prediction band; Phase change material heat flow monitoring module: Monitors the heat flow rate of the component through phase change material patch and heat flow rate sensor, and triggers dynamic compensation when it exceeds a preset threshold; Dynamic compensation algorithm module: recursively updates the compensation coefficients based on historical prediction errors and corrects the input simulation in real time; Closed-loop iteration and threshold adaptation module: Based on the deviation between the compensated predicted band and the actual output, dynamically tighten or loosen the detection domain boundary, and trigger model retraining when the loosening exceeds the limit multiple times. The online optimization module for reinforcement learning updates model parameters in real time based on compensation feedback and state-action-reward design, thereby improving prediction and compensation performance.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the distributed photovoltaic power uncertainty quantification prediction method based on multimodal fusion and adaptive learning as described in any one of claims 1-5.

Images