Photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction

Abstract

The invention provides a photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction. According to the method, a multi-modal data set is formed by collecting photovoltaic array operation and maintenance related data, accurate recognition of a shadow area is achieved through quantization feature enhancement and a dynamic space attention mechanism, shadow boundary coordinates are determined, and the shadow coverage rate is calculated. Further calculating the shadow expansion speed and propagation time according to the shadow boundary coordinates, the solar altitude and the shelter parameters, and predicting the shadow coverage range. Meanwhile, the hot spot risk is quantified through the temperature difference between the shadow area and the non-shadow area and the shadow duration, and graded early warning is triggered. And dynamically adjusting the output voltage of the photovoltaic array based on the shadow coverage rate, and tracking the maximum power point through a particle swarm optimization algorithm. In addition, in combination with a shadow prediction result, the angle of the photovoltaic panel is mechanically adjusted to avoid a shadow area. According to the invention, the power generation efficiency of the photovoltaic array under the shadow shielding condition is improved, and the stability and safety of the system are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent photovoltaic technology, in particular to a photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction. BACKGROUND

[0002] With the increasing global energy demand and the growing emphasis on environmental protection, solar energy as a clean and renewable energy resource has received worldwide attention. Photovoltaic arrays, as an important device for converting solar energy into electrical energy, have been widely used in distributed power generation, large-scale photovoltaic power stations and other fields. However, in actual operation, photovoltaic arrays are often affected by surrounding environmental factors, such as shadowing by trees, buildings and other obstructions, which can cause local shading of photovoltaic cells and a series of problems.

[0003] Specifically, shadowing leads to reduced power generation efficiency, and local shading can change the output current and voltage of photovoltaic cells, resulting in a decrease in the output power of photovoltaic arrays. In severe cases, it may even cause hot spot effect, damaging photovoltaic cells and reducing their service life.

[0004] Currently, traditional shadow detection methods mainly rely on manual inspection or simple sensor monitoring, which have low efficiency, low detection accuracy and cannot predict in real time. For example, manual inspection requires a lot of manpower and time, and it is difficult to find dynamic changes in shadow conditions in time; simple sensor monitoring cannot accurately identify the boundaries and range of shadows, making it difficult to achieve accurate shadow detection.

[0005] Existing photovoltaic arrays lack effective dynamic adjustment mechanisms when facing shadowing. Most photovoltaic systems can only operate in fixed modes and cannot adjust the working state of photovoltaic arrays in real time according to changes in shadow, thus failing to minimize the impact of shadow on power generation efficiency.

[0006] Shadowing can also cause local temperature rise in photovoltaic cells, resulting in hot spot effect. Hot spot effect not only reduces the power generation efficiency of photovoltaic cells, but also accelerates the aging of the cells, and even causes fire hazards and other safety hazards. However, traditional hot spot detection and warning methods are often not timely and accurate enough to effectively prevent the occurrence of hot spot effect. SUMMARY

[0007] The present application aims to at least solve one of the above technical problems in the prior art.

[0008] To this end, the present application provides a photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction.

[0009] The present invention provides a photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction, comprising: Collecting photovoltaic array operation and maintenance related data to form a multimodal data set; the photovoltaic array operation and maintenance related data includes photovoltaic array current, voltage, battery temperature, solar altitude angle, obstruction parameters and infrared thermal value; Based on multimodal datasets, we use quantized feature enhancement and dynamic spatial attention mechanism to identify shadow areas, determine shadow boundary coordinates, and calculate shadow coverage. Calculate the shadow expansion speed and propagation time based on the shadow boundary coordinates, sun altitude angle and obstruction parameters, and predict the shadow coverage range; The temperature difference between shadowed and non-shadowed areas and the duration of shadows are used to quantify the risk of hot spots and trigger graded warnings. Dynamically adjust the photovoltaic array output voltage based on shadow coverage and track the maximum power point through particle swarm optimization algorithm; Based on the shadow prediction results, the angle of the photovoltaic panels is mechanically adjusted to avoid the shadow area.

[0010] The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to the above technical solution of the present invention may also have the following additional technical features: In the above technical solution, based on a multimodal dataset, the shadow area is identified, the shadow boundary coordinates are determined, and the shadow coverage is calculated through quantized feature enhancement and dynamic spatial attention mechanism, including: Acquire image data of the photovoltaic array; The quantized feature extraction method is used to obtain the quantized feature map based on the image data of the photovoltaic array, including:

[0011] in, represents the quantized eigenvalue, i.e. quantum state, the The shadow feature matrix of each channel is used to form a quantized feature map; Indicates the The fusion weight of the quantum states; The first channels; Indicates the A convolution kernel of a quantum state is used to extract the corresponding shadow pattern, each quantum convolution kernel corresponds to a shadow pattern, and the shadow pattern includes horizontal shadow, vertical shadow and edge shadow; Represents the convolution operation; According to the dynamic spatial attention mechanism, a spatial attention weight map is calculated according to the quantized feature map, and the spatial attention weight is used to represent the probability of being a shadow at a corresponding coordinate point, including:

[0012] wherein, represents the spatial attention weight at the coordinate , that is, the probability of being a shadow at the coordinate . represents a Sigmoid activation function; represents a 1x1 convolution weight; represents a feature map height; represents a feature map width; represents a feature map pixel value, that is, a quantized feature value of the first channel; represents an infrared fusion coefficient; represents an infrared thermal value at the coordinate ; The coordinate points with the spatial attention weight greater than a set threshold are counted, and then a shadow coverage rate is calculated, including:

[0013] wherein, represents the shadow coverage rate; represents a spatial attention weight set threshold; By an edge detection algorithm, region edge pixel coordinates satisfying the spatial attention weight greater than the set threshold are extracted from the spatial attention weight map to form a shadow boundary coordinate set.

[0014] In the above technical solution, the image data based on the photovoltaic array adopts a quantum feature extraction method to obtain a quantized feature map, including: Initializing a quantum convolution kernel, so that each quantum convolution kernel corresponds to a shadow mode; Performing quantum kernel convolution on the RGB channels of the input image respectively to obtain a preliminary feature map; Based on the preliminary feature map, multi-channel features are weightedly fused by a quantum state fusion weight to enhance a weak shadow signal.

[0015] In the above technical solution, the dynamic spatial attention mechanism is used to calculate a spatial attention weight map according to the quantized feature map, including: Performing global average pooling on the quantized feature map to extract channel-level global information; Generating an initial spatial weight by 1x1 convolution, and normalizing the initial spatial weight to [0, 1] by a Sigmoid activation function; ​Infrared thermal values ​​are integrated to modify weight distribution and distinguish shadows from interferences.

[0016] In the above technical solution, the calculation of the shadow expansion speed and propagation time based on the shadow boundary coordinates, the sun altitude angle and the obstruction parameters, and the prediction of the shadow coverage range, include: Calculate the shadow expansion speed based on the shadow boundary coordinate set, sun altitude angle and obstruction parameters, including:

[0017] in, Indicates the shadow expansion speed; Represents the shadow boundary coordinate set; Representing coordinates The sun projection vector at ; The calculation method of the solar projection vector includes:

[0018] in, represents the sun's altitude angle; Indicates the horizontal distance of the obstruction; Indicates the coordinates of the base point of the occluder; Calculate the shadow propagation time based on the shadow boundary coordinate set and the shadow expansion speed, including:

[0019] in, represents the shadow travel time; Indicates the center coordinates of the photovoltaic panel.

[0020] In the above technical solution, the hot spot risk is quantified by the temperature difference between the shadow area and the non-shadow area and the shadow duration, and the graded warning is triggered, including: Calculate the temperature difference between the shadow area and the normal area to obtain the basic risk index, including:

[0021]

[0022] in, Indicates the shadow temperature difference; Representing coordinates Battery temperature at Indicates the number of pixels in the shadow area; Indicates the number of pixels in the non-shadow area; represents the basic risk index; Indicates the maximum temperature at which photovoltaic cells can operate safely; Indicates the ambient temperature around the photovoltaic panel; Combining the shadow duration and propagation time, the risk value is amplified by the time weight to obtain a dynamic risk index, including:

[0023] in, represents the dynamic risk index; represents the time influence coefficient; Indicates the current shadow duration; Compare the basic risk index and dynamic risk index with the corresponding set thresholds to determine the current hot spot risk level.

[0024] In the above technical solution, the method of dynamically adjusting the photovoltaic array output voltage based on the shadow coverage ratio and tracking the maximum power point through the particle swarm optimization algorithm includes: When the shadow coverage exceeds the reference value, the voltage compensation amount is calculated. The calculation method of the voltage compensation amount includes:

[0025] in, Indicates the voltage compensation amount; Indicates rated voltage; represents the compensation intensity coefficient; Indicates the current shadow coverage; Indicates the base value of shadow coverage; The particle swarm algorithm is used to dynamically adjust the operating voltage at the next moment based on the difference between the local optimal and global optimal voltages, including:

[0026] in, Indicates the optimal voltage at the next moment; Indicates the current voltage; represents the learning rate; Represents the total number of particles in the particle swarm algorithm; represents the influence coefficient of the jth particle; Indicates the The local optimal voltage searched for each particle; represents the global optimal voltage; Update learning rate and particle weights in real time to improve maximum power point tracking efficiency.

[0027] In the above technical solution, the method of mechanically adjusting the angle of the photovoltaic panel based on the shadow prediction result to avoid the shadow area includes: Calculate the target angle of the photovoltaic panel by combining the height of the obstruction, the sun's altitude angle, and the length of the photovoltaic bracket, including:

[0028] in, Indicates the target angle of the photovoltaic panel; Indicates the height of the obstruction; represents the sun's altitude angle; Indicates the length of the photovoltaic bracket; Indicates the safety margin angle; Based on the current angle and the required angle of the photovoltaic panel, the rotation angle is calculated and the motor rotation command is generated based on the rotation angle, including:

[0029] in, Indicates the rotation angle; Indicates the current angle of the photovoltaic panel; The photovoltaic panel rotates to the target angle based on the rotation command to avoid the shadow.

[0030] The above technical solution also includes: Compare the power output of photovoltaic panels before and after voltage compensation and mechanical adjustment, calculate the efficiency improvement rate, and evaluate the dynamic adjustment results.

[0031] The above technical solution also includes: Based on historical operation data, the quantum state weight is updated by gradient descent method, including:

[0032] in, Indicates the updated The fusion weight of the quantum states; Indicates the parameter update rate; Represents the gradient of the loss function; L is the error between the predicted result and the true value.

[0033] In summary, due to the adoption of the above technical features, the beneficial effects of the present invention are: First, this method collects multimodal data related to photovoltaic array operation and maintenance, including current, voltage, battery temperature, solar altitude, obstruction parameters, and infrared thermal values. It then uses quantized feature enhancement and a dynamic spatial attention mechanism to accurately identify shadow areas. This method can accurately determine the coordinates of shadow boundaries and calculate shadow coverage. Compared to traditional methods, this method significantly improves the accuracy and efficiency of shadow detection, enabling rapid and accurate identification of shadow areas in complex obstruction environments, providing reliable data support for subsequent dynamic adjustments.

[0034] Secondly, the present application can calculate the shadow expansion speed and propagation time according to the shadow boundary coordinates, solar elevation angle and shelter parameters, thereby predicting the shadow coverage range in advance. This prediction function enables the photovoltaic system to take measures in advance to avoid the sudden impact of shadow on power generation efficiency, thereby enhancing the stability and reliability of the system. By predicting the propagation path and range of the shadow in advance, the system can more reasonably plan the layout and adjustment strategy of the photovoltaic panel, thereby minimizing the impact of shadow on power generation efficiency.

[0035] Furthermore, the present application quantifies the hot spot risk by the temperature difference between the shadow area and the non-shadow area and the duration of the shadow, and triggers a hierarchical early warning. This hot spot risk assessment method can timely discover potential hot spot problems and take corresponding early warning measures according to the risk level. This not only helps to protect photovoltaic cells from the damage of hot spot effect and prolong the service life of the cells, but also effectively prevents safety accidents caused by hot spot effect, thereby improving the safety and reliability of the photovoltaic system.

[0036] In addition, the present application dynamically adjusts the output voltage of the photovoltaic array based on the shadow coverage rate, and tracks the maximum power point through the particle swarm optimization algorithm. This method can adjust the working state of the photovoltaic array in real time under the condition of shadow blocking, so as to ensure that the system always operates near the maximum power point, thereby maximizing the power generation efficiency. Compared with the traditional fixed working mode, this dynamic adjustment mechanism can better adapt to the changes of shadow and improve the overall performance and economy of the system.

[0037] Finally, the present application also avoids the shadow area by mechanically adjusting the angle of the photovoltaic panel, thereby further improving the power generation efficiency of the photovoltaic system. The target angle of the photovoltaic panel is calculated based on the height of the shelter, the solar elevation angle and the length of the photovoltaic support, and a motor rotation instruction is generated, so that the photovoltaic panel can be automatically adjusted to the best angle to avoid shadow blocking. This mechanical adjustment measure is combined with the electrical adjustment to realize the all-round optimization of the photovoltaic system, thereby further improving the power generation efficiency and operation stability of the system.

[0038] In summary, the present application realizes the efficient operation of the photovoltaic array under the condition of shadow blocking, improves the power generation efficiency, enhances the stability and safety of the system, and has significant economic and social benefits, thereby providing strong technical support for the development of the photovoltaic industry.

[0039] Additional aspects and advantages of the present application will become apparent from the following description with reference to the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0040] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the following drawings in which: Figure 1is a flow chart of a photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to an embodiment of the present application; Figure 2 is a schematic diagram of photovoltaic panel movable support dynamic adjustment in the photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0042] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other different manners from those described herein, therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0043] The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to some embodiments of the present application will be described below with reference to Figure 1 and Figure 2 .

[0044] Some embodiments of the present application provide a photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction.

[0045] As shown in Figure 1 , the first embodiment of the present application proposes a photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction, including the following steps S1-S6.

[0046] S1, collect photovoltaic array operation and maintenance related data to form a multi-modal data set; the photovoltaic array operation and maintenance related data includes current, voltage, battery temperature, solar elevation angle, shelter parameter and infrared thermal value of the photovoltaic array.

[0047] Specifically, the multi-modal data can be acquired through a sensor network, which can be composed of current sensors, voltage sensors, temperature sensors, solar angle sensors, infrared thermographs, etc., and several kinds of multi-modal data are fused into a multi-modal data set containing time stamps through a data fusion module , wherein, is the output current of the photovoltaic array, i.e. the real-time output current value of the photovoltaic panel, which is used to determine whether the circuit is working normally, and an abnormal value can trigger a short circuit / disconnection warning; is the output voltage of the photovoltaic array, i.e. the real-time output voltage value of the photovoltaic panel, which is used as the core input for power calculation and MPPT (maximum power point tracking) optimization; The surface temperature of the photovoltaic cell, that is, the temperature distribution data on the surface of the photovoltaic panel, can be used for hot spot risk assessment; For the parameters of obstructions (such as trees and buildings), the height of the obstructions and the straight-line distance to the photovoltaic panels were recorded; It is the infrared thermal radiation intensity, that is, the thermal radiation energy per unit area of ​​the photovoltaic panel surface.

[0048] In a specific embodiment, in a large-scale ground photovoltaic power station, data is collected every 100 ms through a distributed sensor network to form a real-time status matrix covering the entire array.

[0049] S2. Based on multimodal datasets, through quantized feature enhancement and dynamic spatial attention mechanism, shadow area recognition is achieved, shadow boundary coordinates are determined, and shadow coverage is calculated.

[0050] In some embodiments, step S2 includes the following steps S21-S25.

[0051] S21. Acquire image data of the photovoltaic array. The image data of the photovoltaic array is a real-time collected surface image of the photovoltaic panel, which serves as an input image for subsequent steps and is usually an RGB image.

[0052] S22. Obtaining a quantized feature map using a quantized feature extraction method based on the image data of the photovoltaic array, including:

[0053] in, represents the quantized eigenvalue, i.e. quantum state, The shadow feature matrix of each channel is used to form a quantized feature map; Indicates the The fusion weight of the quantum states; The first channels, when the input image is an RGB image, Corresponding to R, G, B; Indicates the A 3×3 convolution kernel of quantum states, the number of which is usually 8, is used to extract the corresponding shadow pattern. Each quantum convolution kernel corresponds to a shadow pattern, which includes horizontal shadows, vertical shadows, and edge shadows; Represents a convolution operation.

[0054] In a specific embodiment, the execution process of step S22 is as follows: Initialize 8 quantum convolution kernels so that each quantum convolution kernel corresponds to a shadow pattern; Perform quantum kernel convolution on the RGB channels of the input image to obtain a preliminary feature map; Based on the preliminary feature map, multi-channel features are weightedly fused through quantum state fusion weights to enhance weak shadow signals (such as cloudy shadows).

[0055] S23. Based on the dynamic spatial attention mechanism, a spatial attention weight map is calculated according to the quantized feature map. The spatial attention weight is used to represent the probability that the corresponding coordinate point is a shadow, including:

[0056] in, Representing coordinates The spatial attention weight at coordinate The probability of being a shadow; Represents the Sigmoid activation function; Represents 1×1 convolution weight; Represents the feature map height, that is, the number of vertical pixels of the quantized feature map; Indicates the width of the feature map, that is, the number of horizontal pixels of the quantized feature map; Represents the pixel value of the feature map, that is, the coordinate , No. Quantized eigenvalues ​​of the channel; Indicates the infrared fusion coefficient, which is used to adjust the proportion of infrared thermal value in the attention weight; Representing coordinates Infrared thermal value at.

[0057] In a specific embodiment, the execution process of step S23 is as follows: Perform global average pooling on the quantized feature map to extract channel-level global information; Generate the initial spatial weights through 1×1 convolution, and normalize the initial spatial weights to [0, 1] through the Sigmoid activation function; Based on the low thermal value of shadow areas, the infrared thermal value is integrated and the weight distribution is modified to distinguish shadows from interference objects. Interference objects may be bird droppings, dust, etc.

[0058] S24. Counting the coordinate points whose spatial attention weight is greater than a set threshold, and then calculating the shadow coverage, including:

[0059] in, represents the shadow coverage; Indicates the spatial attention weight setting threshold, which can be set to 0.7.

[0060] Specifically, the shadow coverage rate is mainly used to identify the range of the current single photovoltaic panel covered by the shadow, and the proportion is controlled within the specified range, usually the absolute safety threshold (0%~5%), the early warning threshold (5%~15%), and the dangerous threshold (>15%).

[0061] S25, extracting region edge pixel coordinates satisfying that the spatial attention weight is greater than a set threshold from the spatial attention weight map through an edge detection algorithm to form a shadow boundary coordinate set.

[0062] During overcast or dawn, the contrast between the shadow and non-shadow areas on the surface of the photovoltaic panel is low (the traditional method is prone to missing detection), and the recognition accuracy can be significantly improved by quantum feature enhancement, while the shadow boundary coordinates are accurately extracted. In one specific embodiment, based on the above detection method, the shadow detection accuracy is improved from 82% of the traditional method to more than 98%, the misjudgment rate is reduced from 15% to less than 3%, the extracted shadow boundary coordinates provide key data support for subsequent shadow propagation prediction, and the daily power loss of a single component is reduced by 0.5 kWh.

[0063] S3, calculating the shadow expansion speed and propagation time according to the shadow boundary coordinates, the solar elevation angle and the shelter parameter, and predicting the shadow coverage range.

[0064] In some embodiments, step S3 includes steps S31 and S32.

[0065] S31, calculating the shadow expansion speed according to the shadow boundary coordinate set, the solar elevation angle and the shelter parameter, including:

[0066] wherein, represents the shadow expansion speed, i.e. the moving distance of the shadow edge per unit time; represents the shadow boundary coordinate set; represents the solar projection vector at the coordinate , i.e. the projection direction vector of the sunlight on the photovoltaic panel plane; The calculation method of the solar projection vector includes:

[0067] wherein, represents the solar elevation angle, i.e. the angle between the sunlight and the horizontal plane; represents the horizontal distance of the shelter, i.e. the straight-line distance from the shelter to the photovoltaic panel; represents the shelter base point coordinate, i.e. the pixel coordinate of the center of the bottom of the shelter in the image.

[0068] S32, calculating the shadow propagation time according to the shadow boundary coordinate set and the shadow expansion speed, including:

[0069] wherein, represents the shadow propagation time; represents the center coordinates of the photovoltaic panel.

[0070] In one embodiment, the execution process of step S3 is as follows: calculate the solar projection vector based on the solar elevation angle; combine the horizontal distance of the shelter with the shadow boundary coordinates obtained in step S2 to calculate the moving speed of each point on the shadow edge, i.e. the shadow expansion speed; take the shortest distance from the shadow boundary obtained in step S2 to the center of the photovoltaic panel, divide it by the average shadow expansion speed, and obtain the shadow propagation time.

[0071] In one embodiment, the tree shadow spreads to the inside of the photovoltaic array during the sunrise period (7-9 am). Using the shadow boundary coordinates extracted in step S2, the response time for photovoltaic panel adjustment can be reserved by predicting the shadow expansion speed (e.g. 0.15 m / s) and the propagation time (e.g. 180 s). Specifically, the shadow coverage range can be predicted 30-60 seconds in advance to reduce power fluctuations. When the shadow propagation time is less than 60 seconds, an emergency adjustment is triggered, which can reduce the power loss by 5%-8%.

[0072] S4, quantify the hot spot risk through the temperature difference between the shadow area and the non-shadow area and the shadow duration, and trigger a graded warning.

[0073] In some embodiments, step S4 includes steps S41-S43.

[0074] S41, calculate the temperature difference between the shadow area and the normal area to obtain a basic risk index, including:

[0075]

[0076] wherein, represents the shadow temperature difference, i.e. the average temperature difference between the shadow area and the non-shadow area; represents the cell temperature at coordinates represents the cell temperature at coordinates represents the number of pixels in the shadow area; represents the number of pixels in the non-shadow area; represents the basic risk index, i.e. the quantitative value of the hot spot risk caused by static shadow; represents the maximum temperature for safe operation of the photovoltaic cell; represents the ambient temperature around the photovoltaic panel; S42, combine the shadow duration and the propagation time to amplify the risk value through the time weight to obtain a dynamic risk index, including:

[0077] wherein, represents a dynamic risk index, i.e. a comprehensive risk quantification value considering the shadow duration; represents a time influence coefficient, which can be 0.5; represents the current shadow duration; S43, compare the basic risk index and the dynamic risk index with the corresponding set threshold value respectively, and determine the current hot spot risk level.

[0078] For example, when a cooling measure is triggered, a high-priority warning is triggered.

[0079] In one specific embodiment, when 1 / 4 of the area of the tree branch blocks the monolithic component (for 60s), the shadow temperature difference is calculated to be 20℃, the basic risk index is 1.2, and the dynamic risk index is 1.8, a high-priority warning is triggered. Based on this, the battery piece can be prevented from being burned due to local high temperature (>85℃), and the aging speed of the component can be reduced (the hot spot lasts for 1 hour, which can shorten the service life of the component by 3 months).

[0080] S5, dynamically adjusting the output voltage of the photovoltaic array based on the shadow coverage, and tracking the maximum power point through a particle swarm optimization algorithm.

[0081] In some embodiments, step S5 includes steps S51-S53.

[0082] S51, when the shadow coverage exceeds the reference value, calculating a voltage compensation amount, and the calculation method of the voltage compensation amount includes:

[0083] wherein, represents the voltage compensation amount; represents the rated voltage; represents a compensation intensity coefficient, which can be 0.4; represents the current shadow coverage; represents a reference value of the shadow coverage, i.e. the minimum shadow coverage for starting voltage compensation, which can be 0.2; S52, using a particle swarm algorithm to dynamically adjust the working voltage at the next moment through the difference between the local optimal voltage and the global optimal voltage, including:

[0084] wherein, represents the optimal voltage at the next moment, i.e. the adjusted MPPT working point voltage; represents the current voltage, i.e. the working voltage of the current MPPT; represents the learning rate; Represents the total number of particles in the particle swarm algorithm; represents the influence coefficient of the jth particle; Indicates the The local optimal voltage searched for each particle; represents the global optimal voltage; It should be noted that the specific content of the MPPT algorithm is well known to those skilled in the art and will not be repeated here.

[0085] S53, real-time update of learning rate and particle weight to improve maximum power point tracking efficiency.

[0086] Specifically, when the shadow coverage ratio reaches 25% (exceeding the baseline value of 0.2), the voltage is adjusted from 30V to 32.5V, and combined with MPPT optimization, the power is restored to over 80% of the rated value. Every 1V of voltage compensation increases power output by 3%-5%, and MPPT tracking efficiency is increased from 85% with traditional methods to 95%, resulting in an average daily increase in power generation of approximately 2kWh per module.

[0087] S6. Based on the shadow prediction results, the angle of the photovoltaic panel is mechanically adjusted to avoid the shadow area.

[0088] In some embodiments, step S6 includes the following steps S61-S63.

[0089] S61. Calculate the target angle of the photovoltaic panel based on the height of the obstruction, the sun's altitude angle, and the length of the photovoltaic bracket, including:

[0090] in, Indicates the target angle of the photovoltaic panel; Indicates the height of the obstruction; represents the sun's altitude angle; Indicates the length of the photovoltaic bracket; Indicates the safety margin angle; S62: Calculate the rotation angle based on the current angle and the required angle of the photovoltaic panel, and generate a motor rotation instruction based on the rotation angle, including:

[0091] in, Indicates the rotation angle; Indicates the current angle of the photovoltaic panel; S63: The photovoltaic panel rotates to the target angle based on the rotation instruction to avoid the shadow. Figure 2As shown, the existing photovoltaic support has been able to meet the dynamic adjustment of the photovoltaic panel. The control command is transmitted to the corresponding stepping motor or hydraulic device, so that the photovoltaic panel rotating shaft is adjusted to the target position. It can be understood that the adjustment of the photovoltaic panel is not only through rotation, but also through height adjustment (up and down movement) and position translation (cross movement) and the like.

[0092] For example, when the tree shadow covers the photovoltaic panel, the target angle of the photovoltaic panel is calculated as 25°, the current angle is 12.7°, and the motor is driven to rotate 12.3° to completely avoid the shadow. Through mechanical adjustment, 90% of the light receiving area can be restored, and the daily average power generation of a single component is about 1.2 kWh, and the response time is less than 2 seconds, which is 30 times more efficient than manual adjustment.

[0093] In some embodiments, the photovoltaic array dynamic adjustment method further comprises: Comparing the power output of the photovoltaic panel before and after voltage compensation and mechanical adjustment, calculating the efficiency improvement rate, and evaluating the dynamic adjustment result.

[0094] Specifically, the calculation method of the efficiency improvement rate is as follows:

[0095] Among them, Efficiency improvement rate; Output power after compensation, that is, the actual power generation after optimization; Output power before compensation, that is, the actual power before optimization; The nominal maximum power of the photovoltaic system; In some embodiments, the photovoltaic array dynamic adjustment method further comprises: Based on the real historical operation data, the quantum state weight is updated by gradient descent method, including:

[0096] Among them, The updated fusion weight of the i-th quantum state; The parameter update rate; The loss function gradient; L is the error between the predicted result and the true value. After the system runs for 1 month, the quantum state weight is optimized by real data, so that the shadow detection accuracy is maintained at more than 95%, and the annual power generation can be increased by 5%~10%. At the same time, the model can adapt to seasonal changes (such as the change of shadow mode after the tree leaves fall in winter), and maintain high detection accuracy for a long time.

[0097]

[0098] ​In this specification, illustrative statements about the terminology used do not necessarily limit the scope of the embodiments or examples to a given embodiment or example. Moreover, descriptive terminology such as first, second, etc. is not necessarily used to describe a particular embodiment or example, rather, such terminology is used in accordance with its ordinary meaning to distinguish between two or more instances of an element.

[0099] Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the present application.

Claims

1. A photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction, characterized in that: include: Collecting photovoltaic array operation and maintenance related data to form a multimodal data set; the photovoltaic array operation and maintenance related data includes photovoltaic array current, voltage, battery temperature, solar altitude angle, obstruction parameters and infrared thermal value; Based on multimodal datasets, we use quantized feature enhancement and dynamic spatial attention mechanism to identify shadow areas, determine shadow boundary coordinates, and calculate shadow coverage. Calculate the shadow expansion speed and propagation time based on the shadow boundary coordinates, sun altitude angle and obstruction parameters, and predict the shadow coverage range; The temperature difference between shadowed and non-shadowed areas and the duration of shadows are used to quantify the risk of hot spots and trigger graded warnings. Dynamically adjust the photovoltaic array output voltage based on shadow coverage and track the maximum power point through particle swarm optimization algorithm; Based on the shadow prediction results, the angle of the photovoltaic panels is mechanically adjusted to avoid the shadow area.

2. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 1 is characterized in that: Based on a multimodal dataset, the method uses quantized feature enhancement and a dynamic spatial attention mechanism to identify shadow areas, determine shadow boundary coordinates, and calculate shadow coverage, including: Acquire image data of photovoltaic arrays; The quantized feature extraction method is used to obtain the quantized feature map based on the image data of the photovoltaic array, including: in, represents the quantized eigenvalue, i.e. quantum state, The shadow feature matrix of each channel is used to form a quantized feature map; Indicates the The fusion weight of the quantum states; The first channels; Indicates the A convolution kernel of a quantum state is used to extract the corresponding shadow pattern, each quantum convolution kernel corresponds to a shadow pattern, and the shadow pattern includes horizontal shadow, vertical shadow and edge shadow; Represents the convolution operation; Based on the dynamic spatial attention mechanism, the spatial attention weight map is calculated according to the quantized feature map. The spatial attention weight is used to represent the probability of the corresponding coordinate point being a shadow, including: in, Representing coordinates The spatial attention weight at coordinate The probability of being a shadow; Represents the Sigmoid activation function; Represents 1×1 convolution weight; Indicates the feature map height; Indicates the width of the feature map; Represents the pixel value of the feature map, that is, the coordinate , No. Quantized eigenvalues ​​of the channel; represents the infrared fusion coefficient; Representing coordinates Infrared thermal value at Count the coordinate points whose spatial attention weight is greater than the set threshold, and then calculate the shadow coverage, including: in, represents the shadow coverage; Indicates the threshold for setting spatial attention weight; Through the edge detection algorithm, the edge pixel coordinates of the area where the spatial attention weight is greater than the set threshold are extracted from the spatial attention weight map to form a shadow boundary coordinate set.

3. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 2 is characterized in that: The photovoltaic array-based image data uses a quantized feature extraction method to obtain a quantized feature map, including: Initialize the quantum convolution kernel so that each quantum convolution kernel corresponds to a shadow pattern; Perform quantum kernel convolution on the RGB channels of the input image to obtain a preliminary feature map; Based on the preliminary feature map, multi-channel features are weightedly fused through quantum state fusion weights to enhance weak shadow signals.

4. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 2 is characterized in that: The dynamic spatial attention mechanism is based on which the spatial attention weight map is calculated according to the quantized feature map, including: Perform global average pooling on the quantized feature map to extract channel-level global information; Generate the initial spatial weights through 1×1 convolution, and normalize the initial spatial weights to [0, 1] through the Sigmoid activation function; Infrared thermal values ​​are integrated to modify weight distribution and distinguish shadows from interferences.

5. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 2 is characterized in that: The method of calculating the shadow expansion speed and propagation time based on the shadow boundary coordinates, the sun altitude angle, and the obstruction parameters, and predicting the shadow coverage range, includes: Calculate the shadow expansion speed based on the shadow boundary coordinate set, sun altitude angle and obstruction parameters, including: in, Indicates the shadow expansion speed; Represents the shadow boundary coordinate set; Representing coordinates The sun projection vector at ; The calculation method of the sun projection vector includes: in, represents the sun's altitude angle; Indicates the horizontal distance of the obstruction; Indicates the coordinates of the base point of the occluder; Calculate the shadow propagation time based on the shadow boundary coordinate set and the shadow expansion speed, including: in, represents the shadow travel time; Indicates the center coordinates of the photovoltaic panel.

6. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 1 is characterized in that: The temperature difference between the shadow area and the non-shadow area and the shadow duration are used to quantify the hot spot risk and trigger a graded warning, including: Calculate the temperature difference between the shadow area and the normal area to obtain the basic risk index, including: in, Indicates the shadow temperature difference; Representing coordinates Battery temperature at Indicates the number of pixels in the shadow area; Indicates the number of pixels in the non-shadow area; represents the basic risk index; Indicates the maximum temperature at which photovoltaic cells can operate safely; Indicates the ambient temperature around the photovoltaic panel; Combining the shadow duration and propagation time, the risk value is amplified by the time weight to obtain a dynamic risk index, including: in, represents the dynamic risk index; represents the time influence coefficient; Indicates the current shadow duration; Compare the basic risk index and dynamic risk index with the corresponding set thresholds to determine the current hot spot risk level.

7. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 1 is characterized in that: The method of dynamically adjusting the output voltage of the photovoltaic array based on the shadow coverage ratio and tracking the maximum power point through a particle swarm optimization algorithm includes: When the shadow coverage exceeds the reference value, the voltage compensation amount is calculated. The calculation method of the voltage compensation amount includes: in, Indicates the voltage compensation amount; Indicates rated voltage; represents the compensation intensity coefficient; Indicates the current shadow coverage; Indicates the base value of shadow coverage; The particle swarm algorithm is used to dynamically adjust the operating voltage at the next moment based on the difference between the local optimal and global optimal voltages, including: in, Indicates the optimal voltage at the next moment; Indicates the current voltage; represents the learning rate; Represents the total number of particles in the particle swarm algorithm; represents the influence coefficient of the jth particle; Indicates the The local optimal voltage searched for each particle; represents the global optimal voltage; Update learning rate and particle weights in real time to improve maximum power point tracking efficiency.

8. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 1 is characterized in that: The method of mechanically adjusting the angle of the photovoltaic panel based on the shadow prediction result to avoid the shadow area includes: Calculate the target angle of the photovoltaic panel by combining the height of the obstruction, the sun's altitude angle, and the length of the photovoltaic bracket, including: in, Indicates the target angle of the photovoltaic panel; Indicates the height of the obstruction; represents the sun's altitude angle; Indicates the length of the photovoltaic bracket; Indicates the safety margin angle; Based on the current angle and the required angle of the photovoltaic panel, the rotation angle is calculated and the motor rotation command is generated based on the rotation angle, including: in, Indicates the rotation angle; Indicates the current angle of the photovoltaic panel; The photovoltaic panel rotates to the target angle based on the rotation command to avoid the shadow.

9. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 1, characterized in that: Also includes: Compare the power output of photovoltaic panels before and after voltage compensation and mechanical adjustment, calculate the efficiency improvement rate, and evaluate the dynamic adjustment results.

10. The photovoltaic array dynamic adjustment method based on artificial intelligence shadow detection and prediction according to claim 2, characterized in that: Also includes: Based on historical operation data, the quantum state weight is updated by gradient descent method, including: in, Indicates the updated The fusion weight of the quantum states; Indicates the parameter update rate; Represents the gradient of the loss function; L is the error between the predicted result and the true value.

Images