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

Abstract

The invention discloses a distributed photovoltaic power uncertainty quantitative prediction method based on multi-modal fusion and adaptive learning. The method comprises the following steps: S1, obtaining first accumulated snow coverage data; s2, collecting second accumulated snow influence data; s3, completing multi-modal data fusion in the three-dimensional feature space; s4, in the three-dimensional feature space, an anomaly detection domain is divided according to a preset threshold condition; s5, for a marked region in the anomaly detection domain, generating a power prediction band based on Monte Carlo simulation iteration; s6, activating a dynamic compensation algorithm, calculating a compensation coefficient, and feeding the coefficient back to the power prediction band analog input; and S7, forming an anomaly detection-compensation adjustment-detection domain adaptive closed loop iteration mechanism. According to the method, the field uncertainty such as accumulated snow and hot spots in photovoltaic power prediction is quantified, and the prediction precision and the system response speed are remarkably improved through a dynamic compensation and threshold self-adaptive mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field, and in particular to a distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning. Background Art

[0002] Currently, distributed photovoltaic power plants face significant power generation declines and widening forecast deviations under winter snow cover. Snow not only obscures photovoltaic panels, reducing effective irradiance, but also alters the thermal conductivity of the module surfaces, inducing localized hot spot effects and accelerating module aging or damage. Existing technologies typically use a single data source (such as weather station data or satellite imagery) for power forecasting. This lacks the ability to capture microscopic local thermal anomalies, resulting in delayed response to unusual scenarios (such as sudden snow accumulation) and difficulty in timely error compensation. Furthermore, traditional forecasting methods are often based on static models and cannot dynamically adjust the forecast bandwidth based on real-time environmental changes. This can lead to abnormal power fluctuations that are not reflected in the output power forecast in a timely manner.

[0003] In thermodynamic modeling, current methods mostly ignore the impact of physical mechanisms such as snow crystal structure and phase transitions on the thermodynamic state of photovoltaic panels. In data fusion, isolated use of single-scale (macro or micro) data fails to balance broad coverage with local accuracy. In terms of algorithm adaptability, existing uncertainty quantification methods mostly use fixed parameters and lack the ability to self-correct in real time based on dramatic environmental changes (such as snow melt rate). These shortcomings lead to problems such as wide confidence intervals for power predictions during snow accumulation periods, delayed dynamic compensation mechanisms, and insufficient response to local thermal anomalies in photovoltaic power plants. Summary of the Invention

[0004] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0005] In view of the above-mentioned problems existing in the existing distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning, the present invention is proposed.

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

[0007] To solve the above technical problems, the present invention provides the following technical solution: a distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning, comprising the following steps: S1. Obtain first snow cover data, where the first snow cover data is collected by a satellite remote sensing module and includes shortwave infrared band reflectivity information and corresponding geographic space coordinates; S2. Collecting second snow impact data, where the second snow impact data is collected through an edge computing node array; S3. Perform multimodal data fusion in a three-dimensional feature space, where the X-axis represents snow thickness, the Y-axis represents surface temperature, and the Z-axis represents effective irradiance, and map and associate the first snow cover data with the second snow impact data through interpolation and registration. S4. In the three-dimensional feature space, an anomaly detection domain is divided according to a preset threshold condition, and compensation requirements are automatically marked for data falling into the detection domain; S5. Iteratively generate power prediction bands based on Monte Carlo simulation for the marked areas in the anomaly detection domain to quantify prediction uncertainty; S6. During real-time monitoring, when it is detected that the heat flow rate of the phase change material exceeds a preset threshold, a dynamic compensation algorithm is activated to calculate a compensation coefficient and feed the coefficient back to the power prediction band analog input to correct the prediction result in real time; S7. Dynamically update the abnormality detection domain boundary conditions based on the predicted power deviation after compensation and the actual power output deviation, forming an abnormality detection-compensation adjustment-detection domain adaptive closed-loop iterative mechanism.

[0008] As a preferred solution of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning described in the present invention, wherein: in the step S1, when the integrity of the first snow cover data is lower than the preset quality threshold, it is automatically switched to using only the second snow impact data; otherwise, the two data are fused as described in S3.

[0009] As a preferred solution of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning described in the present invention, the following step is added after step S2: when the surface temperature change rate in the second snow impact data exceeds the preset rate threshold, the edge computing node sampling frequency is automatically increased; otherwise, the original sampling frequency is maintained.

[0010] As a preferred solution of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning described in the present invention, in which: in the step S4, if the number of marked abnormal detection domains exceeds the preset number N continuously, the threshold conditions are lowered in sequence to expand the detection domain, and then step S5 is entered; otherwise, the original threshold is maintained and step S5 is continued.

[0011] As a preferred solution of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning described in the present invention, the threshold conditions include a temperature threshold and a thickness threshold, and 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 domain range.

[0012] As a preferred solution of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning described in the present invention, wherein: in the step S7, if the predicted band deviation and the actual output deviation after compensation show a continuous downward trend, the abnormal detection domain boundary is tightened; otherwise, the abnormal detection domain boundary is relaxed according to a preset ratio; when the detection domain boundary relaxation ratio exceeds the set upper limit after executing step S7 for M consecutive times, the model retraining is triggered to recalibrate the three-dimensional feature space mapping relationship.

[0013] As a preferred solution of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning described in the present 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 with environmental changes.

[0014] A distributed photovoltaic power uncertainty quantitative prediction system based on multimodal fusion and adaptive learning is characterized by including: Satellite remote sensing module: obtains shortwave infrared band reflectivity data and calculates snow cover thickness; Edge computing node array: installed on the surface of each photovoltaic panel, collecting data from temperature sensors, local irradiance sensors and hot spot detection cameras in real time; Three-dimensional feature space construction module: constructs data fusion space based on the snow thickness on the X axis, the component surface temperature on the Y axis, and the effective irradiance on the Z axis; Interpolation and registration module: performs spatial registration and interpolation on macro and micro data to achieve mapping association of data with different resolutions; Abnormal detection domain division module: Based on preset temperature thresholds and thickness thresholds, it automatically identifies power abnormality areas in the feature space and marks the areas that need compensation; Monte Carlo simulation module: runs N random sampling simulations on the marked area to generate power prediction bands; Phase change material heat flow monitoring module: monitors the heat flow rate of the component through phase change material patches and heat flow rate sensors, and triggers dynamic compensation when it exceeds the preset threshold; Dynamic compensation algorithm module: recursively updates the compensation coefficient based on historical prediction errors and corrects the input simulation in real time; Closed-loop iteration and threshold adaptation module: Dynamically tightens or relaxes the detection domain boundary based on the deviation between the compensated prediction band and the actual output, and triggers model retraining when the relaxation exceeds the limit multiple times; Reinforcement learning online tuning module: Based on compensation feedback and state-action-reward design, it updates model parameters in real time to improve prediction and compensation performance.

[0015] Beneficial effects of the present invention: The present invention combines satellite remote sensing with multimodal data from edge computing nodes to effectively capture changes in snow conditions on the surface of photovoltaic modules and in the environment, improving the adaptability of the prediction model to extreme working conditions. Through 3D feature space fusion, 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. A dynamic compensation algorithm based on phase change material heat flow detection enables real-time correction of sudden hot spots or snow accumulation on site; Adaptive threshold and closed-loop iteration mechanism enable the system to automatically adjust the detection domain based on error feedback, improving overall prediction accuracy; Trigger the introduction of retraining and reinforcement learning strategies to ensure continuous model optimization and online adaptability to meet the operational needs of large-scale distributed photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them: Figure 1 Schematic diagram of the steps of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning of the present invention. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0018] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "one embodiment" or "embodiment" 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 various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0020] Furthermore, the present invention is described in detail with reference to schematic diagrams. For ease of illustration, when describing the embodiments of the present invention, cross-sectional views illustrating device structures may be partially enlarged and not to scale. Furthermore, the schematic diagrams are merely illustrative and should not limit the scope of protection of the present invention. Furthermore, in actual production, the three-dimensional dimensions of length, width, and depth should be included.

[0021] Reference Figure 1 , provides a distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning, including the following steps: S1. Obtain first snow cover data, where the first snow cover data is collected by a satellite remote sensing module and includes shortwave infrared band reflectivity information and corresponding geographic space coordinates; S2. Collecting second snow impact data, where the second snow impact data is collected through an edge computing node array; S3. Complete multimodal data fusion in a three-dimensional feature space, where the X-axis represents snow thickness, the Y-axis represents surface temperature, and the Z-axis represents effective irradiance, and map and associate the first snow cover data with the second snow impact data through interpolation and registration; S4. In the three-dimensional feature space, an anomaly detection domain is divided according to a preset threshold condition, and compensation requirements are automatically marked for data falling into the detection domain; S5. For the marked areas in the anomaly detection domain, iteratively generate power prediction bands based on Monte Carlo simulation to quantify prediction uncertainty; S6. During real-time monitoring, when it is detected that the heat flow rate of the phase change material exceeds a preset threshold, a dynamic compensation algorithm is activated to calculate a compensation coefficient and feed the coefficient back to the power prediction band analog input to correct the prediction result in real time; S7. Dynamically update the abnormality detection domain boundary conditions based on the predicted power deviation after compensation and the actual power output deviation, forming an abnormality detection-compensation adjustment-detection domain adaptive closed-loop iterative mechanism.

[0022] Wherein, step S1: obtaining first snow cover data includes: Equipment and modules: Ground reflectivity data is obtained using shortwave infrared (SWIR) band imagers via low-orbit / medium-orbit meteorological satellites deployed above or near photovoltaic power plants.

[0023] Data content: The reflectivity of each pixel in the acquisition area in the 1.5–2.5 μm band and the corresponding geospatial coordinates (longitude, latitude, elevation) are collected. After atmospheric correction and geometric correction, an array of "initial snow cover thickness values" is generated.

[0024] Preprocessing: Temporal registration of multiple satellite data at the same time is performed, and reflectivity is mapped to snow thickness estimates using an empirical model with a resolution of up to 30m×30m.

[0025] Wherein, step S2: collecting the second snow impact data includes: Edge computing node array: An edge node with a temperature sensor (thermocouple or infrared array), irradiance sensor, and GPS coordinate acquisition module is deployed behind or near each group of photovoltaic panels. Each node has local short-term storage and preliminary computing capabilities.

[0026] Sampling content: PV panel surface temperature distribution: obtained through infrared camera (resolution 640×480) or dot matrix temperature sensor; Local irradiance: A high-sensitivity photodiode array is used to measure the actual irradiance on each component; Hot spot position coordinates: The hot spot area is extracted from the temperature distribution image through threshold segmentation, and its center coordinates are recorded.

[0027] Quality control: The node performs time series filtering on the temperature data and occlusion culling (mutation point culling) on ​​the irradiance.

[0028] Specifically, in step S1, when the integrity of the first snow cover data is lower than the preset quality threshold, it is automatically switched to using only the second snow impact data; otherwise, the two data are fused according to S3; the following steps are added after step S2: when the surface temperature change rate in the second snow impact data exceeds the preset rate threshold, the edge computing node sampling frequency is automatically increased; otherwise, the original sampling frequency is maintained.

[0029] Furthermore, step S3: 3D feature space multimodal data fusion includes: Define the coordinate system: Construct a three-dimensional feature space with the X axis representing the snow thickness , the Y axis is the component surface temperature , Z axis is effective irradiance ; Registration and mapping: For each edge node coordinate in the second snow impact data, find its corresponding reflectivity value in the satellite pixel; The snow thickness values ​​of satellite pixels are mapped to the edge node space through bilinear / cubic spline interpolation; Create a mapping function , and construct the fused density function by interpolation in the three-dimensional feature space.

[0030] Among them, in step S4, if the number of marked abnormal detection domains exceeds the preset number N continuously, the threshold conditions are lowered in sequence to expand the detection domain, and then step S5 is entered; otherwise, the original threshold is maintained and step S5 is continued. Specifically, the threshold conditions include temperature threshold and thickness threshold, and when the temperature threshold trigger frequency in step S4 exceeds the threshold trigger frequency, the sensitivity of the thickness threshold judgment is further improved to optimize the abnormal detection domain range.

[0031] Specifically, step S4: anomaly detection domain division and labeling includes: Threshold condition: Set snow thickness threshold in three-dimensional space , temperature threshold , irradiance threshold ; Detection domain: Data points that meet any or a combination of conditions fall into the anomaly detection domain ; Labeling: The system automatically marks internal points as "compensation requirements" and generates corresponding compensation priority sequences; If the number of abnormal domains detected in N consecutive iterations exceeds the preset , then reduce and To expand ; When the temperature threshold trigger frequency exceeds times, increase the sensitivity of the snow thickness threshold (i.e. reduce step size) to optimize the detection domain.

[0032] Wherein, step S5: generating a Monte Carlo simulation quantitative prediction band includes: Initial distribution assumption: Based on the historical observation errors at the same location and under the same meteorological conditions, the probability distribution of power forecast errors is constructed. ; Simulation process: Randomly select 10,000–100,000 groups within the labeled domain Sample, substitute into the photovoltaic module power output model:

[0033] in is the nominal efficiency, 、 is the snow thickness and temperature attenuation coefficient, is the reference temperature; Prediction band: The upper and lower percentiles of the power samples (e.g., 5%–95%) are taken as the prediction band to quantify uncertainty.

[0034] Among them, step S6: dynamic compensation algorithm and real-time correction includes: Phase change material heat flow monitoring: A phase change material (PCM) sensing layer is placed behind the component or at the junction box to monitor the heat flow rate in real time ; Threshold trigger: When When , the dynamic compensation algorithm is triggered and the compensation coefficient is calculated:

[0035] in: is the initial value of the compensation coefficient used in the kth iteration; is the compensation coefficient calculated at the k+1th iteration; is the learning rate, is the actual output, is the median power of the current prediction band; is the actual power output at time t, obtained from field sensors or inverter monitoring data; is the predicted power value at time t, taken from the median (or other statistical indicator, such as mean) of the current prediction band; After each iteration, based on the historical error sequence Recursive Update With initial , ensuring that the compensation coefficient converges quickly to environmental changes.

[0036] Furthermore, in step S7, if the predicted deviation after compensation and the actual output deviation show a continuous downward trend, the anomaly detection domain boundary is tightened; otherwise, the anomaly detection domain boundary is relaxed according to a preset ratio; when the detection domain boundary relaxation ratio exceeds the set upper limit after executing step S7 M times in succession, the model retraining is triggered to recalibrate the three-dimensional feature space mapping relationship.

[0037] Specifically, step S7: closed-loop self-closed iteration includes: Boundary update: According to the prediction error after this compensation Deviation from actual Changing trends: If both show a continuous downward trend, tighten the detection domain boundary and increase , ; Otherwise, according to the preset ratio Relax the boundaries to ensure adequate coverage of outliers.

[0038] Adaptive closed loop: The above detection-compensation-boundary update cycle runs until the prediction accuracy meets the set standard or the maximum number of iterations is reached.

[0039] Specific scenario examples: A large photovoltaic power station in a mountainous area: 1. System deployment and parameter setting Satellite remote sensing module: Select Sentinel-2 satellite SWIR band (1.6 ) data, spatial resolution 20m; Atmospheric correction uses the Sen2Cor tool to generate an array of initial snow thickness values; Empirical Model: Snow Thickness ,in , .

[0040] Edge computing nodes: Deploy 50 nodes, evenly distributed behind a 10MW string module cluster; Temperature sensor: FLIR infrared array, resolution 640×480, temperature accuracy ±0.5℃; Irradiance sensor: Photodiode array, range 0–1500W / m 2 , accuracy ±5W / m 2 ; Phase change material layer sensing: heat flow rate threshold 200W / m 2 .

[0041] 2. Multimodal Fusion and Threshold Conditions Three-dimensional feature space coordinates: Snow thickness ; Surface temperature ; Effective irradiance W / m 2 Initial threshold settings: , , W / m 2 ; 3. Anomaly Detection and Monte Carlo Simulation Detection domain division: When a node measures 、 、 W / m 2 ,fall into , automatically mark compensation needs; If the number of detected domains exceeds 10% for three consecutive days, the trigger threshold will be lowered: , .

[0042] Monte Carlo simulation: Select 50,000 samples and the power model parameters , =0.004 / cm, =0.005 / ; Generate 5%-95% prediction band, such as the middle value of an iteration is predicted to be 8.2MW, the interval is .

[0043] 4. Dynamic compensation and closed-loop iteration Compensation trigger: The heat flow rate is detected at a certain moment , calculate the initial compensation coefficient ; When the actual output is 8.1MW and the predicted median is 8.2MW, the error is -0.1MW, update

[0044] Boundary Update: After 5 iterations, the prediction error is reduced from down to , the detection domain is tightened to , ; Cumulative relaxation ratio , to trigger model retraining; 5. Operation effect Within 30 days, this method reduced the daily forecast root mean square error (RMSE) from 6.5% to 3.1%; For extreme snowy weather (snow thickness > 10 cm), the peak error is reduced from 12% to 5%.

[0045] The present invention also includes a distributed photovoltaic power uncertainty quantification prediction system with multimodal fusion and adaptive learning, including: Satellite remote sensing module: obtains shortwave infrared band reflectivity data and calculates snow cover thickness; Edge computing node array: installed on the surface of each photovoltaic panel, collecting data from temperature sensors, local irradiance sensors and hot spot detection cameras in real time; Three-dimensional feature space construction module: constructs data fusion space based on the snow thickness on the X axis, the component surface temperature on the Y axis, and the effective irradiance on the Z axis; Interpolation and registration module: performs spatial registration and interpolation on macro and micro data to achieve mapping association of data with different resolutions; Abnormal detection domain division module: Based on preset temperature thresholds and thickness thresholds, it automatically identifies power abnormality areas in the feature space and marks the areas that need compensation; Monte Carlo simulation module: runs N random sampling simulations on the marked area to generate power prediction bands; Phase change material heat flow monitoring module: monitors the heat flow rate of the component through phase change material patches and heat flow rate sensors, and triggers dynamic compensation when it exceeds the preset threshold; Dynamic compensation algorithm module: recursively updates the compensation coefficient based on historical prediction errors and corrects the input simulation in real time; Closed-loop iteration and threshold adaptation module: Dynamically tightens or relaxes the detection domain boundary based on the deviation between the compensated prediction band and the actual output, and triggers model retraining when the relaxation exceeds the limit multiple times; Reinforcement learning online tuning module: Based on compensation feedback and state-action-reward design, it updates model parameters in real time to improve prediction and compensation performance.

[0046] The present invention also provides a computer device suitable for a distributed photovoltaic power uncertainty quantification prediction system with 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 implement a distributed photovoltaic power uncertainty quantification prediction system with multimodal fusion and adaptive learning as proposed in the above embodiment.

[0047] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device 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 an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0048] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements a distributed photovoltaic power uncertainty quantification prediction system with multimodal fusion and adaptive learning as proposed in the above embodiment; 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 read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0049] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning, characterized by: The following steps are involved: S1. Obtain first snow cover data, where the first snow cover data is collected by a satellite remote sensing module and includes shortwave infrared band reflectivity information and corresponding geographic space coordinates; S2. Collecting second snow impact data, where the second snow impact data is collected through an edge computing node array; S3. Perform multimodal data fusion in a three-dimensional feature space, where the X-axis represents snow thickness, the Y-axis represents surface temperature, and the Z-axis represents effective irradiance, and map and associate the first snow cover data with the second snow impact data through interpolation and registration. S4. In the three-dimensional feature space, an anomaly detection domain is divided according to a preset threshold condition, and compensation requirements are automatically marked for data falling into the detection domain; S5. Iteratively generate power prediction bands based on Monte Carlo simulation for the marked areas in the anomaly detection domain to quantify prediction uncertainty; S6. During real-time monitoring, when it is detected that the heat flow rate of the phase change material exceeds a preset threshold, a dynamic compensation algorithm is activated to calculate a compensation coefficient and feed the coefficient back to the power prediction band analog input to correct the prediction result in real time; S7. Dynamically update the abnormality detection domain boundary conditions based on the predicted power deviation after compensation and the actual power output deviation, forming an abnormality detection-compensation adjustment-detection domain adaptive closed-loop iterative mechanism.

2. The distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning according to claim 1 is characterized by: In step S1, when the integrity of the first snow cover data is lower than a preset quality threshold, it is automatically switched to using only the second snow impact data; otherwise, the two data are fused as described in S3.

3. The distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning according to claim 2 is characterized by: Add the following step after step S2: when the surface temperature change rate in the second snow impact data exceeds a preset rate threshold, automatically increase the edge computing node sampling frequency; otherwise, maintain the original sampling frequency.

4. The distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning according to claim 1 is characterized by: In step S4, if the number of marked abnormal detection domains exceeds the preset number N consecutively, the threshold conditions are lowered in sequence to expand the detection domain, and then step S5 is entered; Otherwise, the original threshold is maintained and step S5 is executed.

5. The distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning according to claim 4 is characterized by: The threshold conditions include a temperature threshold and a thickness threshold, and when the temperature threshold triggering frequency exceeds the threshold triggering frequency in step S4, the sensitivity of the thickness threshold judgment is further improved to optimize the abnormality detection domain range.

6. The distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning according to claim 1 is characterized by: In step S7, if the predicted deviation after compensation and the actual output deviation show a continuous downward trend, the anomaly detection domain boundary is tightened; otherwise, the anomaly detection domain boundary is relaxed according to a preset ratio; when the detection domain boundary relaxation ratio exceeds the set upper limit after executing step S7 M times in succession, the model retraining is triggered to recalibrate the three-dimensional feature space mapping relationship.

7. The distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning according to claim 1 is characterized by: The dynamic compensation algorithm in step S6 includes adaptive updating of recursive compensation coefficients based on historical prediction errors to ensure that the compensation coefficients converge in real time as the environment changes.

8. A distributed photovoltaic power uncertainty quantitative prediction system based on multimodal fusion and adaptive learning, characterized by: include: Satellite remote sensing module: obtains shortwave infrared band reflectivity data and calculates snow cover thickness; Edge computing node array: installed on the surface of each photovoltaic panel, collecting data from temperature sensors, local irradiance sensors and hot spot detection cameras in real time; Three-dimensional feature space construction module: constructs data fusion space based on the snow thickness on the X axis, the component surface temperature on the Y axis, and the effective irradiance on the Z axis; Interpolation and registration module: performs spatial registration and interpolation on macro and micro data to achieve mapping association of data with different resolutions; Abnormal detection domain division module: Based on preset temperature thresholds and thickness thresholds, it automatically identifies power abnormality areas in the feature space and marks the areas that need compensation; Monte Carlo simulation module: runs N random sampling simulations on the marked area to generate power prediction bands; Phase change material heat flow monitoring module: monitors the heat flow rate of the component through phase change material patches and heat flow rate sensors, and triggers dynamic compensation when it exceeds the preset threshold; Dynamic compensation algorithm module: recursively updates the compensation coefficient based on historical prediction errors and corrects the input simulation in real time; Closed-loop iteration and threshold adaptation module: Dynamically tightens or relaxes the detection domain boundary based on the deviation between the compensated prediction band and the actual output, and triggers model retraining when the relaxation exceeds the limit multiple times; Reinforcement learning online tuning module: Based on compensation feedback and state-action-reward design, it updates model parameters in real time to improve prediction and compensation performance.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the distributed photovoltaic power uncertainty quantitative prediction method based on multimodal fusion and adaptive learning according to any one of claims 1 to 7 are implemented.

Images