Photovoltaic panel angle self-adaptive adjusting system

Abstract

The invention discloses a photovoltaic panel angle adaptive adjustment system, which belongs to the technical field of photovoltaic energy utilization, adopts a three-layer architecture design, and comprises a sensing layer, a decision layer and an execution layer, the sensing layer acquires the operating environment and power generation state data of the photovoltaic panel in real time through a multi-dimensional sensing module for solar attitude, shielding state, component state, irradiance, angle feedback and the like; the decision-making layer comprehensively calculates the optimal adjustment angle of the photovoltaic panel on the basis of a multi-algorithm fusion mechanism in combination with basic angle calculation, multi-factor dynamic correction, machine learning optimization and group control cooperation; the execution layer realizes accurate angle adjustment and reliable system operation through a distributed adjustment control module, a closed-loop correction module and a fault diagnosis module; the system can effectively cope with complex environment changes, avoid shielding influences, optimize heat dissipation and scattered light utilization, improve the overall power generation efficiency through array cooperative control, and has the advantages of being high in self-adaption, high in adjustment precision and remarkable in power generation gain.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic energy utilization technology, and specifically discloses a photovoltaic panel angle adaptive adjustment system. Background Technology

[0002] With the acceleration of the global energy transition, photovoltaic power generation, as an important component of clean and renewable energy, has seen its power generation efficiency optimization become a key research area. The power generation efficiency of photovoltaic panels is directly affected by incident irradiance, which is closely related to the angle of the photovoltaic panel. Ideally, the photovoltaic panel should be aligned with the sun in real time to maximize the reception of direct radiation, thereby improving power output. Therefore, the photovoltaic panel angle adjustment system is one of the core technologies for improving the overall efficiency of photovoltaic systems. Currently, common photovoltaic panel angle adjustment systems are mainly divided into fixed, seasonal manual adjustment, and automatic tracking systems. Fixed systems are simple in structure and low in cost, but cannot adapt to changes in the sun's position, resulting in low power generation efficiency. Seasonal manual adjustment systems require regular manual adjustments, which are inefficient and have a slow response time. Automatic tracking systems achieve angle adjustment through sensors or program control, but still have many limitations. For example, traditional automatic tracking systems are mostly based on a single solar trajectory algorithm, only considering the solar altitude angle and azimuth angle, ignoring environmental factors in actual operation (such as cloud cover, shadows, module temperature, irradiance distribution, etc.), resulting in insufficient adjustment accuracy. In addition, existing systems often lack multi-factor fusion decision-making capabilities and cannot cope with complex weather conditions or shading scenarios, which may lead to adjustment failure or power generation loss. On the other hand, as the scale of photovoltaic arrays expands, the problem of mutual shading between adjacent photovoltaic panels becomes increasingly prominent. Existing systems mostly adopt independent control strategies and lack group control and coordination mechanisms, which can easily lead to a decrease in the overall power generation efficiency of the array. At the same time, traditional systems usually do not have adaptive learning capabilities and cannot optimize adjustment strategies based on historical data and real-time power generation status, resulting in a gradual degradation of the adjustment effect over long-term operation. In addition, the angle feedback and fault diagnosis functions in the execution layer are imperfect, which may lead to the accumulation of adjustment deviations or equipment failures, affecting the reliability of the system. Therefore, it is necessary to invent a photovoltaic panel angle adaptive adjustment system to solve the above problems. Summary of the Invention

[0003] To overcome the aforementioned shortcomings of existing technologies, this invention provides a photovoltaic panel angle adaptive adjustment system, employing a three-layer architecture design, including a perception layer, a decision layer, and an execution layer. The perception layer collects real-time data on the photovoltaic panel's operating environment and power generation status through multi-dimensional perception modules such as solar attitude, shading status, component status, irradiance, and angle feedback. The decision layer, based on a multi-algorithm fusion mechanism, combines basic angle calculation, multi-factor dynamic correction, machine learning optimization, and group control collaboration to comprehensively calculate the optimal adjustment angle of the photovoltaic panel. The execution layer, through distributed adjustment control, closed-loop correction, and fault diagnosis modules, achieves precise angle adjustment and reliable system operation, effectively solving the problems mentioned in the background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a photovoltaic panel angle adaptive adjustment system, comprising a sensing layer, a decision-making layer, and an execution layer; the sensing layer includes a solar attitude sensing module, a shading status sensing module, a component status sensing module, an irradiance sensing module, and an angle feedback sensing module; the decision-making layer includes a basic angle calculation unit, a multi-factor correction unit, a machine learning optimization unit, and a group control and coordination unit. The solar attitude sensing module is configured to combine GPS positioning data, astronomical algorithms and local meteorological station cloud cover data to calculate and correct the real-time solar azimuth and elevation angles. The occlusion state perception module is configured to identify the position, area, and moving speed of the occlusion object based on the improved YOLOv8 image recognition algorithm, and output the occlusion influence coefficient. The irradiance sensing module is configured to separately measure direct irradiance and diffuse irradiance using a high-precision shortwave total radiometer and a matching light-shielding sphere. The multi-factor correction unit is configured to dynamically correct the theoretically optimal light-tracking angle based on the data from the perception layer, and its correction logic follows preset priority rules and superposition restrictions. The group control and coordination unit is configured based on the photovoltaic array topology model. It coordinates the adjustment angle of each photovoltaic panel in the array through conflict detection and resolution strategies to avoid adjacent panels shading each other. The perception layer is used to collect real-time data on the photovoltaic panel's operating environment and power generation status, and output it to the decision layer; the execution layer includes a distributed regulation and control module, a closed-loop correction module, and a fault diagnosis module. Based on the data collected by the perception layer, the decision-making layer calculates the optimal adjustment angle of the photovoltaic panel through the fusion of multiple algorithms and outputs the angle adjustment command to the execution layer. The execution layer receives the angle adjustment command, drives the photovoltaic panel to complete the angle adjustment, and realizes closed-loop correction through angle feedback data to ensure adjustment accuracy.

[0005] Preferably, the component status sensing module collects the back panel temperature of the photovoltaic panel through a distributed temperature sensor, collects the real-time output power, voltage, and current of a single photovoltaic panel through a power sensor, and simultaneously records the cumulative operating time of the component. The angle feedback sensing module collects the actual pitch angle and azimuth angle of each photovoltaic panel in real time and outputs the angle deviation data to the execution layer.

[0006] The basic angle calculation unit calculates the theoretically optimal tracking angle, including the pitch angle θ0 and the azimuth angle φ0, based on the output data of the solar attitude sensing module and by incorporating the solar trajectory equation corrected by the declination angle, hour angle, and geographic latitude. The machine learning optimization unit constructs a prediction model based on random forest, using multi-dimensional data collected by the perception layer as input features and power generation gain adjusted per unit angle as the output target, and optimizes the angle through offline training and online updates.

[0007] Preferably, the solar attitude sensing module calculates the solar trajectory correction coefficient k using the following formula: k = 1 - 0.08 × (C / 10), where C is the cloud cover data from the local meteorological station; and uses the correction coefficient k in combination with the compensation value obtained from on-site calibration to correct the theoretical solar azimuth and altitude angles.

[0008] Preferably, the improved YOLOv8 image recognition algorithm in the shading state perception module is trained using a dataset containing samples of different weather conditions, time periods, and shading types, and the anchor frame parameters are optimized and adapted to the sizes of common shading objects on photovoltaic panels using focal length loss. The shading influence coefficient K is calculated as follows: K=(Sphys / Spanel)×w, where Sphys is the actual shading area calculated through coordinate mapping, Spanel is the light-receiving area of ​​a single photovoltaic panel, and w is a weighting coefficient set according to the type of shading object.

[0009] Preferably, when the solar altitude angle is greater than 5°, the irradiance sensing module operates in a 1-hour cycle, alternating between "30-minute total irradiance measurement" and "30-minute diffuse irradiance measurement", wherein during diffuse irradiance measurement, the shading sphere is adjusted to completely block direct sunlight.

[0010] Preferably, the correction logic of the multi-factor correction unit includes: when the shading influence coefficient is >0.2, adjusting the angle according to the position of the shading object to avoid the shadow area; when the component temperature is >45℃, finely adjusting ±3° based on the theoretical optimal tracking angle to increase the ventilation of the photovoltaic panel surface; when the direct irradiance is <200W / m², adjusting to a fixed optimized angle preset based on historical scattered light power generation data to reduce adjustment energy consumption.

[0011] Preferably, the working logic of the machine learning optimization unit includes: offline training stage: importing environmental-power generation-angle data from the past 12 months to train the initial prediction model; online update stage: comparing the actual power generation after adjustment with the model prediction power every 24 hours, and automatically updating the model parameters if the deviation is >5%; real-time output stage: inputting the angle after multi-factor correction into the optimization model and outputting the optimal adjustment angles θopt and φopt.

[0012] Preferably, the collaborative logic of the group control unit includes: prioritizing the optimal angle of the front row of unobstructed panels; adjusting the angle of the rear row of panels according to the pitch angle of the front row panels to ensure that the lighting surface of the rear row panels is not covered by the shadow of the front row panels; and maximizing the overall power generation gain of the array by iteratively calculating and outputting the final collaborative angle of each panel in the array.

[0013] Preferably, the closed-loop correction module compares the actual angle collected by the angle feedback sensing module with the command angle of the decision layer. If the deviation is >0.5°, it outputs a compensation signal to correct the adjustment amount until the angle deviation is ≤0.5°.

[0014] Preferably, the fault diagnosis module monitors the response speed of angle adjustment and the duration of angle deviation in real time. When the adjustment delay is greater than 2 seconds or the deviation is greater than 1° and lasts for 30 seconds, a fault warning signal is output.

[0015] The technical effects and advantages of this invention are as follows: 1. The system ensures that the photovoltaic panels are always aligned with the sun at a near-optimal angle through high-precision solar trajectory calculation and multi-factor dynamic correction, maximizing the amount of direct irradiance received and thus directly improving power generation efficiency. The machine learning model can learn from massive historical data and real-time operating data, continuously optimize the angle adjustment strategy, and maximize the power generation gain per unit angle adjustment. Compared with systems with fixed angles or simple time-series control, the power generation is significantly improved. 2. By identifying shading and calculating the impact coefficient in real time, the system can actively adjust the angle to avoid the shadow area, effectively solving the problem of "barrel effect" and sudden drop in power generation caused by local shading in traditional arrays; the system can dynamically adjust the strategy according to the actual irradiation conditions and component status to ensure efficient and stable operation under various weather and operating conditions. 3. The group control and coordination unit coordinates the adjustment of the entire photovoltaic array through iterative calculation, avoiding mutual shading between photovoltaic panels, and realizing a leap from "optimal for a single component" to "optimal for the entire array", maximizing the utilization rate of land or roof area and overall power generation revenue. 4. The closed-loop correction mechanism of the execution layer ensures the accuracy of angle adjustment and avoids efficiency loss caused by mechanical errors or external disturbances; the fault diagnosis module can detect abnormalities in the adjustment mechanism in a timely manner and provide early warnings, which facilitates the quick location and handling by operation and maintenance personnel, prevents small faults from evolving into big problems, improves system availability, and reduces maintenance costs and power generation losses. Attached Figure Description

[0016] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0018] Figure 2 This is the multi-factor correction logic diagram of the present invention.

[0019] Figure 3 This is a flowchart of the execution layer control of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides, for example Figure 1 The photovoltaic panel angle adaptive adjustment system shown includes a sensing layer, a decision layer, and an execution layer. The sensing layer includes a solar attitude sensing module, a shading status sensing module, a component status sensing module, an irradiance sensing module, and an angle feedback sensing module. The decision layer includes a basic angle calculation unit, a multi-factor correction unit, a machine learning optimization unit, and a group control and coordination unit. The execution layer includes a distributed adjustment control module, a closed-loop correction module, and a fault diagnosis module.

[0022] The perception layer is used to collect real-time data on the photovoltaic panel's operating environment and power generation status, and output the data to the decision layer. Based on the data collected by the perception layer, the decision-making layer calculates the optimal adjustment angle of the photovoltaic panel through the fusion of multiple algorithms and outputs the angle adjustment command to the execution layer. The execution layer receives the angle adjustment command, drives the photovoltaic panel to complete the angle adjustment, and realizes closed-loop correction through angle feedback data to ensure adjustment accuracy.

[0023] Furthermore, in the above technical solution, the solar attitude sensing module combines GPS positioning data with astronomical algorithms to calculate the real-time azimuth and elevation angles of the sun, and simultaneously accesses local meteorological station cloud cover data to correct the deviation of the solar trajectory. It should be further explained that the detailed execution process of the solar attitude perception module is as follows: 1.1 Obtain GPS positioning data including geographic longitude λ, geographic latitude φ, and UTC timestamp; by sending a request to the meteorological data server, receive and parse the returned data packet, and extract cloud cover data C. Cloud cover data C is a value from 0 to 10, where 0 represents clear sky and 10 represents completely cloudy sky. 1.2 Convert UTC time to local true solar time T, T = UTC time + longitude × 4 minutes / ° + time difference correction; 1.3 Calculate the solar declination angle δ and the hour angle ω: δ≈23.45×sin(360 / 365×(284+n)), where n represents the day of the year in which the current time is, for example, n=32 for February 1st; ω = 15 × (T - 12); 1.4. Combining φ, δ, and ω, calculate the theoretical azimuth angle α and elevation angle h using the solar trajectory equation: h = arcsin[sinφ×sinδ+cosφ×cosδ×cosω], range -90°~90°, negative value indicates the sun is below the horizon; α = arccos[(sinδ×cosφ-cosδ×sinφ×cosω) / cosh], ranging from 0° to 360°, with due south as 0° and west as the positive direction; 1.5. Based on the collected cloud cover data C, calculate the trajectory correction coefficient k using an empirical formula: k = 1 - 0.08 × (C / 10), where C / 10 converts the cloud cover into a coefficient in the range of 0-1. For example, when the cloud cover is 50%, k = 1 - 0.08 × 0.5 = 0.96. The physical meaning of the correction coefficient k is that the more cloud cover there is, the greater the deviation between the actual trajectory of the sun and the theoretical trajectory, and the weight of the theoretical angle needs to be reduced by the coefficient.

[0024] 1.6. Combine the theoretical azimuth angle, elevation angle, and correction coefficients to calculate the final solar attitude data: The actual elevation angle hr = h × k + ho, where ho is a fixed correction value obtained through on-site calibration, typically within the range of ±0.5°. The actual azimuth angle αr = α × k + αo, where αo is a fixed correction value for the azimuth angle, ranging from ±0.5°; It should be further noted that the on-site calibration is performed once per quarter, at noon on a clear, cloudless day, i.e., 12:00±30 minutes local true solar time. The actual solar altitude angle is measured using a professional solar altitude angle measuring instrument, compared with the data output by the module, and ho and αo are adjusted to ensure that the deviation between the corrected azimuth and altitude angles and the actual measured values ​​is ≤0.3°.

[0025] The occlusion state perception module is based on an image recognition algorithm to identify the position, area and moving speed of the occlusion object, and outputs the occlusion influence coefficient in the range of 0-1. It should be further explained that the specific execution process of the occlusion state perception module is as follows: 2.1 Establishing a benchmark template: Acquire 3 frames of standard images of photovoltaic panels under unobstructed conditions, extract the pixel regions of the photovoltaic panel's light-receiving surface using an image segmentation algorithm, determine their coordinate range in the image, and store them as an unobstructed benchmark template for subsequent comparison of obstructed areas; 2.2 Image Recognition Model Construction: An improved YOLOv8 target detection algorithm is adopted, and the following optimizations are made to adapt to photovoltaic shading recognition scenarios: Images of photovoltaic panels were collected under different weather conditions (sunny, cloudy, rainy), at different times of day (early morning, noon, evening), and with different types of occlusion (static occlusion: trees, buildings; dynamic occlusion: birds, clouds; partial occlusion: dust, fallen leaves). The occlusion areas were labeled (boundary boxes + category labels), and a dataset containing more than 100,000 samples was constructed. The dataset was divided into training set, validation set, and test set in an 8:1:1 ratio.

[0026] Model optimization: Simplify the network structure: Remove some redundant convolutional layers from YOLOv8 to reduce computational load and ensure real-time operation on the embedded processor.

[0027] Anchor frame adjustment: Based on the size of the photovoltaic panel and the size of common obstructions (such as birds 5-20cm in size and tree branches 10-50cm in size), re-clustering is performed to generate suitable anchor frame parameters, thereby improving the recognition accuracy of small target obstructions.

[0028] Loss function optimization: Focal loss is introduced to solve the problem of imbalanced samples caused by occlusion and improve the model's generalization ability.

[0029] 2.3 Image Acquisition: The camera captures images of the photovoltaic panel area at preset intervals and records the acquisition timestamps simultaneously; 2.4 Input the acquired image into the YOLOv8 model to identify the position, area, and movement speed of the occlusion, and output the occlusion influence coefficient in the 0-1 range; Position coordinate calculation: Extract the center pixel coordinates (uc, vc) of the bounding box of the obstruction. Combine these coordinates with the camera intrinsic parameters and installation angle, and convert them into physical 3D coordinates (Xc, Yc, Zc) using a coordinate mapping formula, where Zc is the vertical distance between the obstruction and the photovoltaic panel. Output the position identifier of the obstruction within the photovoltaic array.

[0030] It should be further explained that the coordinate mapping formula is: X=(u-u0)×kx×Z / fx, Y=(v-v0)×ky×Z / fy, where (u,v) are pixel coordinates, (X,Y,Z) are physical coordinates, (u0,v0) are principal point coordinates, fx and fy are focal lengths, and kx and ky are pixel physical dimensions.

[0031] Calculation of obstructed area: Pixel area: Calculate the pixel area of ​​the bounding box of the occluder in the image: Spix = (x2 - x1) × (y2 - y1); Physical area conversion: Based on the coordinate mapping relationship, the pixel area is converted into the actual occlusion area Spys, with the formula: Spys = Spix × (kx × ky) × (Zc / fx) 2 ; Movement speed calculation: The Hungarian algorithm is used to associate the trajectory of the same occluder in 5 consecutive frames of images to obtain the center coordinate sequence {(X1, Y1, t1), (X2, Y2, t2), ..., (X5, Y5, t5)}; Speed ​​calculation: The horizontal movement speed v is calculated through linear fitting, using the following formula: ; Where ΔX and ΔY are the displacement changes within 5 frames, i.e., ΔX=X5-X1 and ΔY=Y5-Y1, and Δt is the time interval, Δt=t5-t1.

[0032] Calculation of shading impact coefficient: The shading impact coefficient K takes into account both the shading area ratio and the type of shading object. The formula is: K=(Sphys / Spanel)×w; where Spanel is the light-receiving area of ​​a single photovoltaic panel, usually 1.6m²; w is the weighting coefficient of the shading object. Preferably, static hard objects such as trees have w=1.0, dynamic soft objects such as birds have w=0.3, and semi-transparent shadows have w=0.5. Data output: Structured data is output every 100ms, including: category, location coordinates, occlusion area (Sphys), movement speed (v), and occlusion impact coefficient (K); The component status sensing module collects the back panel temperature of the photovoltaic panel through distributed temperature sensors, collects the real-time output power, voltage and current of a single photovoltaic panel through power sensors, and simultaneously records the cumulative running time of the component. In a preferred embodiment of the present invention, each photovoltaic panel is equipped with two DS18B20 digital temperature sensors. The temperature sensor cables are concealed along the photovoltaic panel support to avoid direct sunlight. A high-precision power sensor is selected, integrating a voltage transformer, a current transformer, and a power calculation module to support real-time measurement of active power, RMS voltage, and RMS current. The power sensor is installed in a dedicated slot in the combiner box, with a distance of ≥5cm from other electrical components to prevent electromagnetic interference. The irradiance sensing module collects the incident irradiance of the photovoltaic panel's light-collecting surface through a high-precision shortwave total radiation meter, distinguishing between direct irradiance and diffuse irradiance. It should be further explained that the high-precision shortwave total radiometer is equipped with a manual shading ball, which, through a bracket linked to the radiometer, can block direct sunlight to measure scattered irradiance separately; the specific implementation of the irradiance sensing module is as follows: Furthermore, the voltage signal output by the high-precision shortwave total radiation meter is collected every 3 seconds and converted into the original irradiance value. The calculation formula is irradiance = voltage value × 200W / m² / mV, where irradiance is the total irradiance. When the solar altitude angle is >5°, the direct / scattered irradiance separation process is initiated; when it is <5°, only the total irradiance is output without separation. The separation cycle is 1 hour, and it is executed alternately with "30 minutes of total irradiance measurement and 30 minutes of scattered irradiance measurement" to ensure the timeliness and stability of the data.

[0033] Furthermore, during the scattering measurement period, the position of the shading sphere is adjusted so that the line connecting the center of the shading sphere and the center of the radiation meter's sensing surface is aligned with the direction of the sun. The shading range must completely cover the radiation meter's sensing surface, and there must be no light leakage at the edges. Whether there is light leakage is verified by real-time monitoring of the irradiance value. After shading, the irradiance must be reduced to less than 30% of the irradiance before shading. After the shading state stabilizes, that is, the irradiance remains stable for 10 seconds, irradiance data is collected at a 3-second cycle to obtain the scattered irradiance. Direct irradiance calculation: The diffuse irradiance measured in the previous cycle is used, according to the formula: Direct irradiance = Total irradiance - Diffuse irradiance; The angle feedback sensing module collects the actual pitch angle and azimuth angle of each photovoltaic panel in real time and outputs the angle deviation data to the execution layer.

[0034] It should be further explained that the specific execution process of the angle feedback sensing module is as follows: Each photovoltaic panel is equipped with a dual-axis tilt sensor, with a measurement range of pitch angle 0°-90° and azimuth angle 0°-360°, and a measurement accuracy of ±0.1°; The angle feedback sensing module obtains the current target angle issued by the decision-making layer, including the target pitch angle and the target azimuth angle, and calculates the angle deviation based on the current pitch angle and azimuth angle of the photovoltaic panel.

[0035] Furthermore, in the above technical solution, the basic angle calculation unit calculates the theoretically optimal tracking angle, including the optimal theoretical pitch angle θ0 and the optimal theoretical azimuth angle φ0, based on the output data of the solar attitude sensing module and by incorporating the solar trajectory equation corrected by declination angle, hour angle, and geographic latitude. The multi-factor correction unit dynamically corrects the theoretically optimal tracking angle based on the output data of the shading state sensing module, component state sensing module, and irradiance sensing module. The machine learning optimization unit constructs a prediction model based on random forest, using multi-dimensional data collected by the sensing layer as input features and power generation gain per unit angle adjustment as the output target, and optimizes the angle through offline training and online updates. The group control and coordination unit coordinates the adjustment angles of each photovoltaic panel in the array based on the photovoltaic array topology model to avoid mutual shading between adjacent panels.

[0036] It should be further explained that the specific execution process of the basic angle calculation unit is as follows: Receive real-time solar azimuth angle αr, altitude angle hr, declination angle δ, hour angle ω, and cloud cover data C from the solar attitude sensing module; simultaneously receive local geographic parameters (latitude φ, longitude λ) and true solar time T; Theoretical pitch angle θ 0c calculate: The theoretical tilt angle of a photovoltaic panel needs to complement the solar altitude angle to maximize the reception of direct sunlight. The formula is: θ 0c =arcsin[sinφ×sinδ+cosφ×cosδ×cosω]; (Derivation logic: By using the spherical trigonometric relationship, the normal of the photovoltaic panel is aligned with the direction of sunlight).

[0037] Result correction: When the solar altitude angle hr < 5°, force θ0 = 5° to avoid frequent operation of the adjustment mechanism.

[0038] Theoretical azimuth φ 0c calculate: The theoretical azimuth angle of a photovoltaic panel needs to be aligned with the sun's direction. The formula is: φ 0c =arccos[(sinδ×cosφ-cosδ×sinφ×cosω) / coshr]; (Azimuth definition: due south is 0°, west is positive, range 0°~360°).

[0039] Cloud cover correction and result optimization: Corrected trigger condition: When the solar attitude module outputs cloud cover data, enable the cloud cover correction factor k.

[0040] Corrected formula: θ0 = θ 0c ×k+θos;φ0=φ0c ×k+φos; where θos and φos are empirical compensation values ​​obtained through calibration using historical data; Every 10 seconds, the multi-factor correction unit outputs the calculated results, including the corrected theoretical pitch angle and theoretical azimuth angle.

[0041] Furthermore, in the above technical solution, the correction logic of the multi-factor correction unit includes: when the shading influence coefficient is >0.2, adjusting the angle according to the position of the shading object to avoid the shadow area; when the component temperature is >45℃, finely adjusting ±3° based on the theoretical optimal tracking angle to increase the ventilation of the photovoltaic panel surface; when the direct irradiance is <200W / m², adjusting to a fixed optimized angle preset based on historical scattered light power generation data to reduce adjustment energy consumption.

[0042] It should be further explained that the specific execution process of the multi-factor correction unit is as follows: Initial angle acquisition: The theoretical optimal pitch angle θ0 and azimuth angle φ0 output by the basic angle calculation unit are used as the initial angle values.

[0043] Occlusion correction (highest priority): If the occlusion influence coefficient K > 0.2: Based on the coordinates of the shading object's location, the overlap ratio between the shaded area and the photovoltaic panel's light-receiving surface is calculated.

[0044] Adjust the angle: Correct the azimuth or pitch angle in the direction that avoids the shadow. The correction range is dynamically adjusted according to the proportion of the shaded area (maximum not exceeding 15°) to ensure that the light-receiving surface of the photovoltaic panel is out of the shadow area.

[0045] Example: If the obstruction is located to the east of the photovoltaic panel, the azimuth angle is increased by 5-10° from φ0; if the obstruction is located above, the pitch angle is decreased by 3-8°.

[0046] Temperature correction: If the component temperature T > 45℃: Determine heat dissipation requirements based on irradiance data: When direct irradiance is >500W / m², prioritize increasing ventilation volume and correct the pitch angle towards +3°; when direct irradiance is <500W / m², correct towards -3°.

[0047] Corrected angle: θtemp = θ0 ± 3°, ± is determined according to the heat dissipation direction, and the azimuth angle remains unchanged at φ0.

[0048] Irradiance correction: If direct irradiance < 200 W / m²: Instead of real-time light tracking, a preset diffused light angle is used to reduce the energy consumption of the adjustment mechanism.

[0049] If the direct irradiance is ≥200W / m² for 5 consecutive cycles, it will automatically switch back to real-time correction mode and restore the dynamic correction angle through 3 smooth transitions (each correcting 1 / 3 of the deviation).

[0050] Correcting priority and stacking rules: Priority: Shading correction > Temperature correction > Irradiance correction, with high-priority correction results used as initial values ​​for low-priority corrections; Overlay restrictions: If the angle deviation after overlaying occlusion correction and temperature correction is >10°, then occlusion correction will take precedence, and the temperature correction will be halved. Furthermore, the correction logic of the multi-factor correction unit is as follows: Figure 2 As shown; The corrected angle is output to the machine learning optimization unit every 10 seconds.

[0051] Furthermore, in the above technical solution, the working logic of the machine learning optimization unit includes: offline training stage: importing environmental-power generation-angle data from the past 12 months to train the initial prediction model; online update stage: comparing the actual power generation after adjustment with the model prediction power every 24 hours, and automatically updating the model parameters if the deviation is >5%; real-time output stage: inputting the angle after multi-factor correction into the optimization model and outputting the optimal adjustment angles θopt and φopt.

[0052] It should be further explained that the specific execution process of the machine learning optimization unit is as follows: During the offline training phase, either upon initial system deployment or manually triggered: Data set preparation: Import 12 months of historical environmental-power generation-angle data, including fields such as solar azimuth angle, altitude angle, shading influence coefficient, module temperature, direct / scattered irradiance, corrected angle, and actual power generation, with a data volume of ≥100,000 records.

[0053] Data cleaning: Outliers were removed, and missing values ​​were supplemented using linear interpolation. The training set and test set were divided in a 7:3 ratio.

[0054] Feature engineering: The input features are normalized (mapped to the 0-1 interval), and features strongly correlated with power generation gain (correlation ≥ 0.3) are selected by Pearson correlation coefficient, and finally 12 core features are retained.

[0055] Model training: Set the random forest model parameters as follows: n_estimators=100, max_depth=15, min_samples_split=2, min_samples_leaf=1, bootstrap=True (sampling rate 70%).

[0056] Using 12-dimensional environmental and angular features as input, and the power generation gain (ΔP / Δθ) per unit angle adjustment as the output target, the model is trained on the training set for 100 iterations, with the training error recorded in each iteration.

[0057] Model evaluation and storage: Evaluate the model performance using the test set, calculate the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²), requiring R² ≥ 0.85.

[0058] If the evaluation criteria are met, the model is saved to an SSD, and the training parameters, evaluation results, and timestamps are recorded. It is then marked as a "production-ready model" using MLflow.

[0059] During the online update phase, it will be executed once every 24 hours: Data collection and processing: Collect real-time operating data from the past 24 hours, including environmental data, corrected angles, and actual power generation, with a data volume of ≥1000 records, and process the data according to the characteristic format of offline training.

[0060] Deviation calculation: The predicted power generation is obtained by predicting the corrected angle over the past 24 hours using the currently deployed model. This predicted power generation is then compared with the actual power generation collected by the component status sensing module to calculate the average deviation.

[0061] Model update decision: If the average bias is >5%, start incremental training: add the newly collected data to the training set, maintain a training / test ratio of 7:3, freeze some of the underlying parameters of the model, update only the weights of the top decision tree, and iterate for 50 rounds.

[0062] If the average deviation is ≤5%, the model is not updated; only the current performance index is recorded as a reference for subsequent optimization.

[0063] Validation and deployment after update: Re-evaluate the test set with the updated model. If R² ≥ 0.85 and the deviation ≤ 5%, mark it as the new version via MLflow and deploy it online; otherwise, abandon the update and retain the original model.

[0064] During the real-time optimization phase, it is executed every 10 seconds: Input data acquisition and preprocessing: Receive the corrected angle output by the multi-factor correction unit and simultaneously collect 12-dimensional real-time feature data (solar attitude, occlusion coefficient, etc.) of the perception layer.

[0065] Preprocess the real-time feature data: normalize to the 0-1 range, remove instantaneous outliers (deviation from the average of the previous 5 times > 30%), and generate the model input vector.

[0066] Optimal Angle Prediction: Substitute the preprocessed input vector into the random forest model to predict the power generation gain under different angle adjustment schemes, and select the angle corresponding to the maximum power generation gain as the optimal adjustment angle (θopt, φopt).

[0067] Angle constraints: The optimal angle must be within the mechanical adjustment range (pitch angle 0°-90°, azimuth angle 0°-360°). If it exceeds the range, the nearest boundary value is taken as the final angle.

[0068] Furthermore, in the above technical solution, the collaborative logic of the group control collaborative unit includes: prioritizing the optimal angle of the front row of unobstructed panels; adjusting the angle of the rear row of panels according to the pitch angle of the front row panels to ensure that the lighting surface of the rear row panels is not covered by the shadow of the front row panels; and maximizing the overall power generation gain of the array by iteratively calculating and outputting the final collaborative angle of each panel in the array.

[0069] It should be further explained that the specific execution process of the group control and collaboration unit is as follows: Receive the optimal angles θopt and φopt of a single board from the machine learning optimization unit; synchronously acquire the actual angles θact and φact from the angle feedback sensing module; Bind the optimal angle, actual angle, and equipment status according to the photovoltaic panel ID to generate a single device adjustment requirement: Δθ=θopt-θact, Δφ=φopt-φact; Filtering invalid requirements: When |Δθ| < 0.5° and |Δφ| < 0.5°, it is determined as "no adjustment required"; Occlusion conflict detection: For the adjustment requirements Δθ and Δφ of each board, the potential shading of adjacent boards is analyzed in conjunction with topological relationships: Pitch angle conflict: If θact(A) ≥ θact(B) + 5° after adjustment of board A, and B is the board directly behind A, it is judged as a longitudinal occlusion risk; Azimuth conflict: If the deviation between φact(A) and φact(C) after adjustment of plate A is >10°, and C is the adjacent plate to the side of A with a distance <1.5m, it is judged as a risk of lateral obstruction; Generate a conflict matrix: record the conflict source board ID, affected board ID, and risk level; Conflict resolution strategies: Priority sorting: Sort by "number of affected boards × risk level", and handle high-priority conflicts first.

[0070] Furthermore, the risk levels are divided into three levels: high, medium, and low. The core definitions and judgment thresholds for each level are as follows: High risk: Number of affected panels ≥ 2, obstruction area > 20%, predicted duration > 10 minutes; Medium risk: Number of affected panels = 1, obstruction area accounts for 5%-20%, predicted duration is 5-10 minutes; Low risk: Number of affected panels = 1, obstruction area < 5%, predicted duration < 5 minutes; Angle correction: Corrects the optimal angle of the conflict source board, for example: Longitudinal occlusion risk: θopt(A) is modified to θact(B)+4°, with a 1° safety margin.

[0071] Lateral occlusion risk: φopt(A) is corrected by 3° in the direction of φact(C) to reduce the angle difference.

[0072] Load balancing control: The cumulative adjustment time of the statistical execution agency will be used to postpone the adjustment of equipment that has been running continuously for more than 20 seconds until the number of idle equipment is ≥5.

[0073] Adjustments should be made in batches according to region, with each batch adjusting ≤6 boards to avoid overloading the power supply circuit.

[0074] Standardized final angle commands are generated for adjustment requirements based on conflict detection and load assessment.

[0075] Furthermore, in the above technical solution, the distributed adjustment and control module receives the final angle command output by the decision-making layer and assigns an independent control signal to each photovoltaic panel; the closed-loop correction module compares the actual angle collected by the angle feedback sensing module with the command angle of the decision-making layer. If the deviation is >0.5°, it outputs a compensation signal to correct the adjustment amount until the angle deviation is ≤0.5°; the fault diagnosis module monitors the response speed of the angle adjustment and the duration of the angle deviation in real time. When the adjustment delay is >2s or the deviation is >1° and lasts for 30s, it outputs a fault warning signal.

[0076] Further explanation is needed regarding the specific implementation of the photovoltaic panel angle adaptive adjustment system execution layer as follows: The execution layer includes a distributed adjustment control module, a closed-loop correction module, and a fault diagnosis module. These modules work together to achieve precise angle adjustment and reliable system operation. In terms of hardware deployment, each photovoltaic panel is equipped with an independent servo controller and a DC geared motor. The servo controller supports pulse + direction control mode. The closed-loop correction module reuses the dual-axis tilt sensor of the angle feedback sensing module. The fault diagnosis module is equipped with a current sensor, a time counter, and an angle deviation detector. The main controller integrates an audible and visual alarm module and supports sending fault signals to the operation and maintenance platform via Ethernet. During installation, the servo controller and motor are fixed to the side of the photovoltaic panel bracket, with a distance of ≥20cm from the photovoltaic panel backplate. Power cables and signal cables are laid separately with a spacing of ≥10cm. All shielded cables are grounded. Each device is affixed with an ID label consistent with the photovoltaic panel. The main controller pre-stores an ID and communication address mapping table.

[0077] In practice, the distributed regulation and control module first receives the final angle command issued by the decision-making group control and coordination unit, which includes the photovoltaic panel ID, target pitch angle, azimuth angle, and adjustment priority. Each servo controller filters and parses the command according to its own ID, converts the target angle into a pulse signal to drive the motor. During the adjustment process, the motor current is monitored. If the current is >3A, the speed is automatically reduced to 2° / s. If the current is continuously >4A, the adjustment is paused and an early warning is issued. The adjustment status is fed back to the main controller in real time. The closed-loop correction module collects the actual angle through the tilt sensor and calculates the angle deviation (Δθ, Δφ) every 5 seconds. When the total deviation Δtot (Δtot=Δθ+Δφ) is greater than 0.5°, a compensation command is generated. The compensation angle is Δtot×1.2. The drive motor makes a fine adjustment at a speed of 1° / s. This process is repeated until the deviation is ≤0.5°. If the correction fails to meet the standard after 3 consecutive corrections, a fault diagnosis is triggered. The fault diagnosis module monitors the adjustment delay, angle deviation, and motor current in real time. When the adjustment delay is greater than 2 seconds, it is determined to be a response speed fault. When the angle deviation is greater than 1° and lasts for 30 seconds, it is determined to be an adjustment accuracy fault. After a fault occurs, the subsequent adjustment of the photovoltaic panel is suspended. If the fault is not resolved after 10 minutes, the motor power supply is cut off. During regular maintenance, the adjustment accuracy is calibrated quarterly using a high-precision electronic tilt meter at a typical angle. The servo controller response delay is tested and optimized annually. The overload current threshold is adjusted every six months according to the motor status to ensure long-term stable operation of the system.

[0078] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A photovoltaic panel angle adaptive adjustment system, characterized in that, It includes the perception layer, decision-making layer, and execution layer; The perception layer includes a solar attitude perception module, an occlusion state perception module, a component state perception module, an irradiance perception module, and an angle feedback perception module; the decision-making layer includes a basic angle calculation unit, a multi-factor correction unit, a machine learning optimization unit, and a group control and coordination unit; the execution layer includes a distributed adjustment and control module, a closed-loop correction module, and a fault diagnosis module. The solar attitude sensing module is configured to combine GPS positioning data, astronomical algorithms and local meteorological station cloud cover data to calculate and correct the real-time solar azimuth and elevation angles. The occlusion state perception module is configured to identify the position, area, and moving speed of the occlusion object based on the improved YOLOv8 image recognition algorithm, and output the occlusion influence coefficient. The irradiance sensing module is configured to separately measure direct irradiance and diffuse irradiance using a high-precision shortwave total radiometer and a matching light-shielding sphere. The multi-factor correction unit is configured to dynamically correct the theoretically optimal light-tracking angle based on the data from the perception layer, and its correction logic follows preset priority rules and superposition restrictions. The group control and coordination unit is configured based on the photovoltaic array topology model. It coordinates the adjustment angle of each photovoltaic panel in the array through conflict detection and resolution strategies to avoid adjacent panels shading each other. The perception layer is used to collect real-time data on the photovoltaic panel's operating environment and power generation status, and output the data to the decision layer. Based on the data collected by the perception layer, the decision-making layer calculates the optimal adjustment angle of the photovoltaic panel through the fusion of multiple algorithms and outputs the angle adjustment command to the execution layer. The execution layer receives the angle adjustment command, drives the photovoltaic panel to complete the angle adjustment, and realizes closed-loop correction through angle feedback data to ensure adjustment accuracy.

2. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: The component status sensing module collects the back panel temperature of the photovoltaic panel through distributed temperature sensors, collects the real-time output power, voltage and current of a single photovoltaic panel through power sensors, and simultaneously records the cumulative running time of the component. The angle feedback sensing module collects the actual pitch angle and azimuth angle of each photovoltaic panel in real time and outputs the angle deviation data to the execution layer. The basic angle calculation unit calculates the theoretically optimal tracking angle, including the pitch angle θ0 and the azimuth angle φ0, based on the output data of the solar attitude sensing module and by incorporating the solar trajectory equation corrected by the declination angle, hour angle, and geographic latitude. The machine learning optimization unit constructs a prediction model based on random forest, using multi-dimensional data collected by the perception layer as input features and power generation gain adjusted per unit angle as the output target, and optimizes the angle through offline training and online updates.

3. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: The solar attitude sensing module calculates the solar trajectory correction coefficient k using the following formula: k = 1 - 0.08 × (C / 10), where C is the cloud cover data from the local meteorological station; and uses the correction coefficient k in combination with the compensation value obtained from on-site calibration to correct the theoretical solar azimuth and altitude angles.

4. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: The improved YOLOv8 image recognition algorithm in the shading state perception module is trained using a dataset containing samples of different weather conditions, time periods, and shading types. It also uses focal length loss to optimize and adapt the anchor frame parameters to the sizes of common shading objects on photovoltaic panels. The shading influence coefficient K is calculated as follows: K = (Sphys / Spanel) × w, where Sphys is the actual shading area calculated through coordinate mapping, Spanel is the light-receiving area of ​​a single photovoltaic panel, and w is a weighting coefficient set according to the type of shading object.

5. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: When the solar altitude angle is greater than 5°, the irradiance sensing module operates in a 1-hour cycle, alternating between "30-minute total irradiance measurement" and "30-minute diffuse irradiance measurement". During diffuse irradiance measurement, the shading sphere is adjusted to completely block direct sunlight.

6. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: The correction logic of the multi-factor correction unit includes: when the shading influence coefficient is greater than 0.2, the angle is adjusted according to the position of the shading object to avoid the shadow area; when the component temperature is greater than 45°, the angle is finely adjusted by ±3° based on the theoretical optimal tracking angle to increase the ventilation of the photovoltaic panel surface; when the direct irradiance is less than 200W / m², the angle is adjusted to a fixed optimized angle preset based on historical scattered light power generation data to reduce the adjustment energy consumption.

7. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: The working logic of the machine learning optimization unit includes: offline training phase: importing environmental-power generation-angle data from the past 12 months to train the initial prediction model; online update phase: comparing the actual power generation after adjustment with the model's predicted power every 24 hours, and automatically updating the model parameters if the deviation is >5%; real-time output phase: inputting the angle after multi-factor correction into the optimization model and outputting the optimal adjustment angles θopt and φopt.

8. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: The collaborative logic of the group control unit includes: prioritizing the optimal angle of the front row of unobstructed panels; adjusting the angle of the rear row of panels according to the pitch angle of the front row panels to ensure that the lighting surface of the rear row panels is not covered by the shadow of the front row panels; and maximizing the overall power generation gain of the array by iteratively calculating and outputting the final collaborative angle of each panel in the array.

9. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: The closed-loop correction module compares the actual angle collected by the angle feedback sensing module with the command angle of the decision layer. If the deviation is greater than 0.5°, it outputs a compensation signal to correct the adjustment amount until the angle deviation is less than or equal to 0.5°.

10. The photovoltaic panel angle adaptive adjustment system as described in claim 1, characterized in that: The fault diagnosis module monitors the response speed of angle adjustment and the duration of angle deviation in real time. When the adjustment delay is greater than 2 seconds or the deviation is greater than 1° and lasts for 30 seconds, it outputs a fault warning signal.

Images