Solar irradiance prediction method, system and device, and electronic equipment

Abstract

The invention relates to the field of atmospheric remote sensing, and provides a solar irradiance prediction method, system and device, and electronic equipment, and the method comprises the steps: obtaining cloud bottom image data of a target cloud body collected by a target unmanned plane; performing attitude correction on the cloud bottom image data, and performing semantic segmentation to obtain a binary cloud body mask sequence; performing feature extraction on the binary cloud body mask sequence; and inputting the extracted feature vector into a pre-trained regression model to obtain a solar irradiance prediction result below the flight path of the target unmanned aerial vehicle. The method, the device and the system are used for solving the technical blank that high-temporal-spatial-resolution and foresight local solar irradiance prediction on a front route cannot be provided for a moving unmanned aerial vehicle in the prior art, and according to the scheme, through a visual sensor carried by the unmanned aerial vehicle, microscopic dynamic characteristics of a cloud base are quantitatively analyzed in real time; and a quantitative relation between the characteristics and a ground irradiance slope event is established, so that active early warning of the flight path risk is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of atmospheric remote sensing, and in particular to a solar irradiance prediction method and system, device and electronic equipment. BACKGROUND

[0002] The rapid and severe fluctuation of solar irradiance, i.e. ramp events, poses a serious challenge to multiple fields. In the field of renewable energy, irradiance ramp events can cause the output power of photovoltaic power stations to fluctuate dramatically, impacting the stability of the power grid. For an increasing number of solar-powered long-endurance unmanned aerial vehicles, a sudden decrease in irradiance (entering cloud shadow) on the flight path can cause a sudden decrease in energy input, which, if the battery energy storage is insufficient, will directly threaten flight safety. In addition, for unmanned aerial vehicles performing tasks such as agricultural remote sensing and mapping, dramatic changes in light conditions can severely affect the quality and consistency of image data.

[0003] In related technologies, there are some technologies for irradiance prediction, such as ground-based sky imagers (GSI), meteorological satellites, and airborne point sensors, etc. The GSI system can effectively analyze the cloud conditions above by observing the sky with a fisheye camera, but it is a fixed station and can only provide point prediction at its location, and cannot provide irradiance prediction for the front path of an unmanned aerial vehicle moving in a wide area. In addition, there is a difference between the observation angle of the cloud and the unmanned aerial vehicle flying at the cloud bottom, making it difficult to accurately depict the key cloud features affecting the unmanned aerial vehicle; the spatial resolution of the satellite cloud image of the meteorological satellite is usually hundreds of meters to kilometers, and the temporal resolution is several minutes to tens of minutes. This spatio-temporal scale cannot meet the needs of capturing and analyzing the internal microstructure of the cloud (such as edge morphology and fragmentation process) that causes ramp events of irradiance in seconds to minutes.

[0004] In summary, there is currently a lack of a technical means that can enable an unmanned aerial vehicle to "actively perceive" the cloud dynamics on the front route and accordingly predict the impending irradiance fluctuation in real time and prospectively. SUMMARY

[0005] The application provides a solar irradiance prediction method and system, device and electronic equipment, to solve the technical blank that the related art cannot provide a high spatio-temporal resolution and forward-looking local solar irradiance prediction on a front route for a moving unmanned aerial vehicle, the scheme of the application quantitatively analyzes micro dynamic characteristics of a cloud bottom in real time through a visual sensor carried by the unmanned aerial vehicle, and establishes a quantitative relationship between the characteristics and a ground irradiance slope event, so that active early warning of a flight path risk is realized.

[0006] The application provides a solar irradiance prediction method, comprising: obtaining cloud bottom image data of a target cloud body collected by a target unmanned aerial vehicle; After posture correction is performed on the cloud bottom image data, semantic segmentation is performed to obtain a binary cloud body mask sequence; feature extraction is performed on the binary cloud body mask sequence; a feature vector obtained through the extraction is input into a pre-trained regression model to obtain a solar irradiance prediction result below a flight path of the target unmanned aerial vehicle.

[0007] According to the solar irradiance prediction method provided by the application, the posture correction performed on the cloud bottom image data comprises: obtaining an intrinsic matrix of an image acquisition module of the target unmanned aerial vehicle and a rotation matrix of the target unmanned aerial vehicle; based on the intrinsic matrix and the rotation matrix, a homography transformation matrix corresponding to the cloud bottom image data is calculated; based on the homography transformation matrix, perspective transformation is performed on the cloud bottom image data to obtain cloud bottom image data after posture correction.

[0008] According to the solar irradiance prediction method provided by the application, the semantic segmentation to obtain a binary cloud body mask sequence comprises: the cloud bottom image data after the posture correction is input into a pre-trained semantic segmentation network model, and the binary cloud body mask sequence is output; the semantic segmentation network model is a full convolutional neural network model.

[0009] According to the solar irradiance prediction method provided by the application, the feature extraction performed on the binary cloud body mask sequence comprises: calculation is performed on the binary cloud body mask sequence to obtain cloud body motion characteristics, cloud body morphological characteristics and cloud body evolution characteristics of the target cloud body; the cloud body motion characteristics comprise a motion speed and a motion direction of the target cloud body; the cloud body morphological characteristics comprise a cloud amount and an edge density of the target cloud body; The cloud body evolution feature includes a fragmentation rate of the target cloud body, and the fragmentation rate represents a rate of change of a total length of an edge of the target cloud body.

[0010] According to the solar irradiance prediction method provided by the application, the semantic segmentation network model adopts a Dice loss function for model training, and conforms to the following formula:

[0011] Wherein, is a probability that the i th pixel predicted by the model is a cloud, is a true value.

[0012] According to the solar irradiance prediction method provided by the application, the training process of the regression model includes: Determine the real-time horizontal projection distance of the target cloud body based on the cloud body shape feature and the cloud body evolution feature of the target cloud body; Determine the time delay of the target cloud body based on the motion speed and the real-time horizontal projection distance of the target cloud body; The sum of the time delay and the prediction length is used as the model input, and the solar irradiance prediction result after the prediction length is used as the model output, and the regression model is trained.

[0013] According to the solar irradiance prediction method provided by the application, after obtaining the solar irradiance prediction result under the flight path of the target unmanned aerial vehicle, the method further includes: Compare the solar irradiance prediction result with a preset safety threshold; If the solar irradiance prediction result is lower than the safety threshold, a safety warning is performed.

[0014] The application also provides a solar irradiance prediction system, comprising: A data acquisition module is configured to acquire cloud bottom image data of a target cloud body collected by a target unmanned aerial vehicle; A semantic segmentation module is configured to perform semantic segmentation on the cloud bottom image data after posture correction to obtain a binary cloud mask sequence; A feature extraction module is configured to extract features from the binary cloud mask sequence; An irradiance prediction module is configured to input the extracted feature vector into a pre-trained regression model to obtain a solar irradiance prediction result under the flight path of the target unmanned aerial vehicle.

[0015] The application also provides a solar irradiance prediction device installed on a target unmanned aerial vehicle, comprising: A data processing module is configured to perform any of the above solar irradiance prediction methods; An image acquisition module comprising at least one camera for continuously acquiring an image sequence of the target cloud above the flight path of the target UAV; A pose perception module for acquiring the speed and attitude of the target UAV; A wireless communication module for data communication with the ground station.

[0016] The application also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the solar irradiance prediction method according to any one of the above when executing the program.

[0017] The application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the solar irradiance prediction method according to any one of the above.

[0018] The application also provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the solar irradiance prediction method according to any one of the above.

[0019] In the solar irradiance prediction method provided by the application, the cloud bottom image data of the target cloud in the target area can be automatically acquired by the UAV, after the data is attitude-corrected to eliminate the influence of the attitude jitter of the UAV, semantic segmentation and feature extraction can be performed, and then the processed feature vector can be input into a pre-trained regression model to obtain the corresponding solar irradiance prediction result, so that the UAV changes from passive acceptance of illumination changes to active prediction and avoidance. For a solar-powered UAV, the flight safety (risk of crashing due to energy depletion) under partial cloudy weather can be reduced by at least 50%. For optical remote sensing tasks, the path can be planned in advance to avoid cloud shadows, reducing the proportion of invalid data collection by 20%-40%, in addition, the super-high resolution irradiance prediction with second-level refresh and meter-level perception is provided, which can accurately capture rapid slope events caused by small-scale clouds, in addition, the UAV carrying the application can not only complete its own task, but also becomes a mobile advanced atmospheric detection platform. The quantitative cloud feature data collected and processed by the UAV can serve the power generation prediction of ground photovoltaic power stations, or be used for scientific research of small-scale meteorological phenomena, and create new data service value. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0021] Figure 1 is a flowchart of a solar irradiance prediction method provided by an embodiment of the present application; Figure 2 is a structural diagram of a solar irradiance prediction system provided by an embodiment of the present application; Figure 3 is a physical structure diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0023] Figure 1 is a flowchart of a solar irradiance prediction method provided by an embodiment of the present application.

[0024] As shown in Figure 1 , the present embodiment provides a solar irradiance prediction method, comprising: Step 101, obtaining cloud bottom image data of a target cloud body collected by a target unmanned aerial vehicle; In actual application, the target unmanned aerial vehicle can fly to a target area where solar irradiance prediction is needed to perform image collection. Specifically, the cloud bottom image data sequence of the target cloud body in the target area can be continuously captured at a preset frequency.

[0025] Step 102, after posture correction of the cloud bottom image data, performing semantic segmentation to obtain a binary cloud body mask sequence; In actual application, the posture correction of the cloud bottom image data comprises: obtaining an intrinsic matrix of an image collection module of the target unmanned aerial vehicle and a rotation matrix of the target unmanned aerial vehicle; calculating a homography transformation matrix corresponding to the cloud bottom image data based on the intrinsic matrix and the rotation matrix; performing perspective transformation on the cloud bottom image data based on the homography transformation matrix to obtain the cloud bottom image data after posture correction.

[0026] Specifically, the purpose of attitude correction is to remove the influence of the UAV's own motion, especially attitude changes, on the image, generating a sequence of cloud images with a "space-based fixed viewpoint". During the attitude correction process, the camera's intrinsic parameter matrix K can be obtained through pre-calibration. When capturing the i-th frame, the UAV's attitude angles (roll angle, pitch angle, yaw angle) are synchronously obtained from the pose perception module. Based on this, the rotation matrix from the UAV's body coordinate system to the ground horizontal coordinate system is calculated. .

[0027] Rotation matrix It is achieved by adjusting the drone's attitude angle, i.e., roll angle. (roll), pitch angle (pitch) and yaw angle (yaw) is a combination of a series of basic rotation operations. This process follows the standard method for converting Euler angles to rotation matrices.

[0028] In practice, the rotation from the aircraft coordinate system to the ground horizontal coordinate system follows a specific rotation sequence. In this embodiment, yaw, pitch, and roll rotations can be performed sequentially, specifically including the following steps: 1. Yaw around the Z-axis (Yaw, Rotate to obtain the rotation matrix. :

[0029] 2. Pitch around the Y-axis. Rotate to obtain the rotation matrix. :

[0030] 3. Roll around the X-axis (Roll, Rotate to obtain the rotation matrix. :

[0031] Final rotation matrix It is the product of the three basic rotation matrices mentioned above. According to the rotation order of ZYX, its calculation formula is as follows: .

[0032] Combining camera intrinsic parameters K and rotation matrix Calculate the homography transformation matrix to correct a tilted image to a horizontal viewpoint. The calculation formula is as follows:

[0033] Applying homography matrix inverse of the matrix The original image Imgi is subjected to a perspective transformation (WarpPerspective) to obtain a posture-corrected image:

[0034] After this processing, all images Img' in the sequence are as if taken from a fixed, horizontally upward-facing camera, eliminating the effects of UAV posture jitter.

[0035] Further, after posture correction of the cloud bottom image data, semantic segmentation can also be performed, which refers to inputting the posture-corrected cloud bottom image data into a pre-trained semantic segmentation network model to output the sequence of binary cloud mask. The semantic segmentation network model applied in this embodiment can be a U-Net semantic segmentation network model, i.e., a fully convolutional neural network model, which includes a compression path (encoder) for capturing context information and a symmetric expansion path (decoder) capable of achieving precise positioning. The encoder is stacked by multiple convolutional layers and maximum pooling layers, and the decoder is composed of up-sampling convolution, feature map splicing (Skip Connection) with the encoder, and convolutional layers. This structure can effectively fuse deep semantic features and shallow detail features, and achieve precise segmentation of cloud edges.

[0036] When training the semantic segmentation network model, thousands of UAV skyward-view cloud bottom images under different weather conditions can be collected and manually pixel-level labeled to generate binary truth masks of clouds (foreground) and clear sky (background), and then the model can be trained using the Dice loss function (Dice Loss). Compared with cross-entropy loss, Dice loss can better handle the scenario where the proportion of cloud and clear sky pixels is uneven, and conforms to the following formula:

[0037] wherein, is the probability that the i-th pixel predicted by the model is a cloud, is the true value.

[0038] The Adam optimizer can be used subsequently, with an initial learning rate of 1e -4 , and the Early Stopping strategy is used to prevent model overfitting, ultimately obtaining a high-performance model specialized for cloud bottom image segmentation.

[0039] Step 103, feature extraction is performed on the sequence of binary cloud masks; The feature extraction on the sequence of binary cloud masks includes: The cloud motion feature of the target cloud body includes a motion speed and a motion direction of the target cloud body. The cloud motion feature of the target cloud body includes a motion speed and a motion direction of the target cloud body. The cloud shape feature of the target cloud body includes a cloud amount and an edge density of the target cloud body. The cloud evolution feature of the target cloud body includes a fragmentation rate of the target cloud body, and the fragmentation rate represents a change rate of a total length of an edge of the target cloud body.

[0040] In the extraction of the cloud motion feature, a RAFT dense optical flow algorithm can be used to calculate a pixel displacement field (u_{pix}, v_{pix}) between two continuous corrected cloud codes M_t and M_{t-1}, and then ground displacement (Delta x_{uav}, Delta y_{uav}) of the UAV within Delta t time is obtained, and the ground displacement is converted into pixel displacement (Delta u_{uav}, Delta v_{uav}) according to the flight height H of the UAV and the camera field of view FOV. In the implementation, the average pixel motion vector of the target cloud body is the difference between the apparent motion and the UAV motion, which satisfies the following formula:

[0041] Then, the pixel speed (u_{cloud}, v_{cloud}) can be converted into physical speed (unit: meter / second) and direction according to the height H and the camera parameters.

[0042] In the extraction of the cloud shape feature, the cloud amount can be calculated by the following formula:

[0043] Wherein, M_t(i,j) is the value (0 or 1) of the mask image at pixel (i,j) at time t, W and H_{img} are the image width and height; The edge density of the target cloud body can be calculated by the following formula:

[0044] Wherein, represents the edge contour obtained by applying the Canny edge detection algorithm to the cloud code M_t, and Length calculates the total length (pixel number) thereof; In actual application, the fragmentation rate can also be defined by using the change rate of the total length of the cloud edge to quantify the evolution intensity of the cloud: .

[0045] Step 104, input the extracted feature vector into the pre-trained regression model to obtain the prediction result of the solar irradiance under the target UAV flight path.

[0046] In implementation, the regression model can use XGBoost (Extreme Gradient Boosting) model, which has excellent performance in feature combination and high computational efficiency, suitable for embedded deployment.

[0047] XGBoost is an ensemble learning model based on gradient boosting decision tree (GBDT). It continuously fits the residual of the previous round of model by iteratively adding new decision trees, and finally adds the prediction results of all trees to obtain the final output. The input of the regression model in this embodiment is the multi-dimensional time series feature vector Xt constructed in the above steps, and the output is a scalar, i.e. the predicted future irradiance value Yt.

[0048] In practical application, the core of the regression model in predicting solar irradiance is to consider the time delay Delta t_{lag} of the cloud from being observed by the UAV to its shadow projection to the ground station. Specifically, during model training, the delay can be calculated according to the real cloud speed V_{cloud}(t) calculated at time t and the real-time horizontal projection distance D_{proj}(t) between the UAV and the ground station: Δt lag (t)= D proj (t) / V cloud (t) At each time t, an input feature vector Xt is constructed, which contains not only the features at the current time, but also the features at the previous N times to provide time series information: Xt=[Ft,Ft-1,...,Ft-N], where Ft=[Ca(t),De(t),Rf(t),...].

[0049] The ground station then adds the irradiance value at time t + Delta t_{lag}(t) + a prediction time Delta t_{pred} (e.g. 30 seconds):

[0050] The final training dataset consists of a large number of (X_t, Y_t) sample pairs. The collected (X_t, Y_t) dataset is divided into a training set (80%) and a test set (20%) in chronological order. Key hyperparameters of XGBoost (such as the number of trees n_estimators, the maximum tree depth max_depth, and the learning rate) are optimized using methods such as cross-validation and grid search. With the optimal hyperparameter configuration, the model is trained using the training set, aiming to minimize the root mean square error (RMSE) between the predicted and true values.

[0051] The provided solar irradiance prediction method can automatically acquire cloud base image data of target cloud bodies in the target area through UAVs. After attitude correction of the data to eliminate the influence of UAV attitude jitter, semantic segmentation and feature extraction can be performed. Then, the processed feature vector can be input into a pre-trained regression model to obtain the corresponding solar irradiance prediction result, enabling UAVs to transform from passively receiving changes in light to actively predicting and avoiding them.

[0052] In an exemplary embodiment, after obtaining the predicted solar irradiance below the flight path of the target UAV, the method further includes: Compare the predicted solar irradiance results with a preset safety threshold; If the predicted solar irradiance is lower than the safety threshold, a safety warning will be issued.

[0053] For example, a safety warning could be "Warning: Expected to enter an area with 70% irradiance reduction in 35 seconds." The drone can then make risk avoidance decisions based on this warning information.

[0054] The solar irradiance prediction system provided by the present invention is described below. The solar irradiance prediction system described below can be referred to in correspondence with the solar irradiance prediction method described above.

[0055] Figure 2 This is a schematic diagram of the solar irradiance prediction system provided in an embodiment of the present invention.

[0056] like Figure 2 As shown, the solar irradiance prediction system provided in this embodiment includes: Data acquisition module 201 is used to acquire cloud base image data of the target cloud body collected by the target UAV; The semantic segmentation module 202 is used to perform semantic segmentation on the cloud base image data after attitude correction to obtain a binarized cloud mask sequence. Feature extraction module 203 is used to extract features from the binarized cloud mask sequence; The irradiance prediction module 204 is configured to input the extracted feature vector into a pre-trained regression model to obtain a prediction result of solar irradiance under a flight path of the target UAV.

[0057] In an example embodiment, the semantic segmentation module 202 is further configured to: obtain an intrinsic matrix of an image acquisition module of the target UAV and a rotation matrix of the target UAV; calculate a homography transformation matrix corresponding to the cloud base image data based on the intrinsic matrix and the rotation matrix; perform perspective transformation on the cloud base image data based on the homography transformation matrix to obtain cloud base image data after pose correction.

[0058] In an example embodiment, the semantic segmentation module 202 is further configured to: input the cloud base image data after pose correction into a pre-trained semantic segmentation network model to output the sequence of binary cloud masks; The semantic segmentation network model is a fully convolutional neural network model.

[0059] In an example embodiment, the feature extraction module 203 is further configured to: perform calculation on the sequence of binary cloud masks to obtain cloud motion features, cloud shape features, and cloud evolution features of the target cloud; The cloud motion features include a motion speed and a motion direction of the target cloud; The cloud shape features include a cloud amount and an edge density of the target cloud; The cloud evolution features include a fragmentation rate of the target cloud, and the fragmentation rate represents a rate of change of a total length of an edge of the target cloud.

[0060] In an example embodiment, the semantic segmentation module 202 is further configured to: The semantic segmentation network model adopts a Dice loss function for model training, and conforms to the following formula:

[0061] wherein, is a probability of the i-th pixel being cloud predicted by the model, is a true value.

[0062] In an example embodiment, the training process of the regression model includes: determining a real-time horizontal projection distance of the target cloud based on the cloud shape features and the cloud evolution features of the target cloud; determine a time delay of the target cloud body based on a moving speed and a real-time horizontal projection distance of the target cloud body; train the regression model by taking a sum of the time delay and a prediction duration as a model input and taking a solar irradiance prediction result after the prediction duration as a model output.

[0063] In an example embodiment, a safety warning module is further included, and specifically configured to: compare the solar irradiance prediction result with a preset safety threshold; if the solar irradiance prediction result is lower than the safety threshold, perform safety warning.

[0064] The specific implementation method of the solar irradiance prediction system provided in the embodiment can be implemented with reference to the above-described embodiments, which will not be described herein again.

[0065] The scheme of the present application further provides a solar irradiance prediction device installed on a target unmanned aerial vehicle, comprising: a data processing module configured to perform the solar irradiance prediction method described in any of the above embodiments; an image acquisition module comprising at least one camera configured to continuously acquire an image sequence of a target cloud body above a flight path of the target unmanned aerial vehicle; the camera can be a high-definition, high-dynamic-range (HDR) wide-angle camera installed towards the sky, configured to continuously capture an image sequence of a cloud bottom above the unmanned aerial vehicle at a frame rate not lower than 5 Hertz (Hz).

[0066] a pose perception module configured to acquire a speed and an attitude of the target unmanned aerial vehicle; the pose perception module is integrated with a global positioning system (GPS) and an inertial measurement unit (IMU) configured to acquire an accurate geographic position, speed and attitude (roll, pitch and yaw) of the device in real time, and the data of the device is used for motion compensation and geographic registration of the captured images.

[0067] a wireless communication module configured to perform data communication with a ground station; specifically, the wireless communication module is configured to perform bidirectional data communication with the ground station or a cloud server, on one hand, to receive real-time high-frequency irradiance data from the ground station (for online calibration or offline training of the model), and on the other hand, to send the calculated irradiance prediction result to a flight control system of the unmanned aerial vehicle.

[0068] The following describes a specific embodiment of the solar irradiance prediction method and system and device provided by the present application.

[0069] The present embodiment describes the application of the device of the present application on a solar unmanned aerial vehicle with a wingspan of 4 meters. The unmanned aerial vehicle performs routine inspection tasks at a height of 300 meters above a large ground photovoltaic power station, and the cloud bottom is about 1000 meters high.

[0070] The unmanned aerial vehicle is provided with an image acquisition module, including an industrial-grade 4K camera, configured with a 120° wide-angle lens, and capable of capturing images at 10 fps; the data processing module is an NVIDIA Jetson AGX Orin embedded computer, and a high-precision pyranometer is arranged in the central area of the photovoltaic power station to record the global horizontal irradiance (GHI) at a frequency of 10 Hz, and the data are communicated in real time with the unmanned aerial vehicle through a 5G network.

[0071] In the model training phase, the unmanned aerial vehicle flies for 1 hour over the power station for calibration. The device synchronously records the cloud bottom image sequence and the ground station irradiance data. The data processing module performs offline calculation after the flight, extracts the complete cloud feature time sequence, and aligns it with the irradiance sequence, trains an XGBoost regression model, which can predict the irradiance value in the next 30 seconds according to the current time cloud features.

[0072] In the application process, at the beginning, for example, at t=0, the unmanned aerial vehicle flies at a constant speed along the preset route. The cloud layer above it is relatively thin and broken. Real-time analysis by the device shows that the average cloud speed is 8 m / s, the direction is basically consistent with the route, the edge density is high, the normalized value is 0.8, and the fragmentation rate is moderate, the value is 0.5. These features are input into the XGBoost model, and the prediction result is: “the irradiance in the next 30 seconds is stable, and the fluctuation is less than ±50 W / m²”. The flight control system receives this information and continues to fly along the original route.

[0073] At t=120s, a large cumulus cloud with clear edges and thick clouds appears in front of the unmanned aerial vehicle route. The camera of the device captures the cloud and starts analysis. U-Net completely segments it, and the optical flow algorithm measures its moving speed as 9 m / s, with a 30° angle between the direction and the route, and the edge density feature value drops sharply to 0.2, and the fragmentation rate approaches 0, indicating that a stable and strong shading cloud is approaching.

[0074] After these features are input into the model, an immediate warning is output: “serious slope event warning! It is predicted that 25 seconds later, it will enter the strong cloud shadow area, and the irradiance will drop from 850 W / m² to below 200 W / m², lasting about 90 seconds.” The flight control system of the unmanned aerial vehicle receives this quantitative warning. Its energy management system determines that if it enters the cloud shadow according to the original route, the net consumption rate of the battery will exceed the safety threshold. The flight control system immediately starts the path re-planning algorithm to calculate a detour route that is offset 300 meters to the side, which can perfectly avoid the predicted cloud shadow core area.

[0075] The unmanned aerial vehicle executes the new route and successfully bypasses the cloud shadow, and the on-board irradiance sensor shows that the power input is stable throughout the process. Without the device of the present application, the unmanned aerial vehicle would enter the strong cloud shadow according to the original route, and it is highly likely that it would be forced to make an emergency landing due to insufficient energy, resulting in a failed mission.

[0076] The solar irradiance prediction method provided by the above embodiments has the following beneficial effects: 1. Active and forward-looking risk avoidance is achieved: the unmanned aerial vehicle changes from passive acceptance of light changes to active prediction and avoidance. For solar unmanned aerial vehicles, flight safety (risk of crashing due to energy depletion) under partial cloudy weather can be reduced by at least 50%. For optical remote sensing tasks, the path can be planned in advance to avoid cloud shadows, reducing the proportion of invalid data collection by 20-40%.

[0077] 2. Unprecedented spatiotemporal resolution is provided: super-high resolution irradiance prediction with second-level refresh and meter-level perception is provided, accurately capturing rapid slope events caused by small and medium-sized clouds, which cannot be achieved by any existing satellite or ground observation network.

[0078] 3. Expanding the application value of unmanned aerial vehicles: unmanned aerial vehicles not only complete their own tasks, but also become a mobile advanced atmospheric detection platform. The quantitative cloud feature data collected and processed by the unmanned aerial vehicles can serve power generation prediction for ground photovoltaic power stations or scientific research of small-scale meteorological phenomena, creating new data service value.

[0079] Figure 3 An example of an entity structure diagram of an electronic device is shown in Figure 3 The electronic device can include a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communications bus 340. The processor 310 can invoke logical instructions in the memory 330 to execute a solar irradiance prediction method, which includes: Obtaining cloud bottom image data of a target cloud body collected by a target unmanned aerial vehicle; After attitude correction of the cloud bottom image data, performing semantic segmentation to obtain a binary cloud mask sequence; Feature extraction is performed on the binary cloud mask sequence; Inputting the extracted feature vector into a pre-trained regression model to obtain a solar irradiance prediction result under the flight path of the target unmanned aerial vehicle.

[0080] In addition, the logic instructions in the memory 330 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the method of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0081] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program is executed by a processor, so that the computer can execute the solar irradiance prediction method provided by the above-mentioned methods, and the method comprises: obtaining cloud bottom image data of a target cloud collected by a target unmanned aerial vehicle; After the attitude correction of the cloud bottom image data, semantic segmentation is performed to obtain a binary cloud mask sequence; feature extraction is performed on the binary cloud mask sequence; the extracted feature vector is input into a pre-trained regression model to obtain a solar irradiance prediction result below the flight path of the target unmanned aerial vehicle.

[0082] In another aspect, the present application also provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the solar irradiance prediction method provided by the above-mentioned methods, and the method comprises: obtaining cloud bottom image data of a target cloud collected by a target unmanned aerial vehicle; After the attitude correction of the cloud bottom image data, semantic segmentation is performed to obtain a binary cloud mask sequence; feature extraction is performed on the binary cloud mask sequence; the extracted feature vector is input into a pre-trained regression model to obtain a solar irradiance prediction result below the flight path of the target unmanned aerial vehicle.

[0083] The device embodiments described above are merely illustrative, wherein the units illustrated as separate components can or can not be physically separated, and the components illustrated as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0084] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and include a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of the embodiments or some parts of the embodiments.

[0085] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting solar irradiance, characterized in that, include: Acquire cloud base image data of the target cloud body collected by the target drone; After attitude correction of the cloud base image data, semantic segmentation is performed to obtain a binarized cloud mask sequence; Feature extraction is performed on the binarized cloud mask sequence; The extracted feature vectors are input into a pre-trained regression model to obtain the predicted solar irradiance below the target UAV's flight path.

2. The solar irradiance prediction method according to claim 1, characterized in that, The attitude correction of the cloud base image data includes: Obtain the intrinsic parameter matrix of the image acquisition module of the target UAV and the rotation matrix of the target UAV; Based on the intrinsic parameter matrix and the rotation matrix, calculate the homography transformation matrix corresponding to the cloud base image data; Based on the homography transformation matrix, perspective transformation is performed on the cloud base image data to obtain cloud base image data after attitude correction.

3. The solar irradiance prediction method according to claim 1, characterized in that, The semantic segmentation process, which yields a binary cloud mask sequence, includes: The pose-corrected cloud base image data is input into a pre-trained semantic segmentation network model, and the output is the binarized cloud mask sequence. The semantic segmentation network model is a fully convolutional neural network model.

4. The solar irradiance prediction method according to claim 1, characterized in that, The feature extraction of the binarized cloud mask sequence includes: The binarized cloud mask sequence is calculated to obtain the cloud motion characteristics, cloud morphology characteristics, and cloud evolution characteristics of the target cloud. The cloud motion characteristics include the target cloud's speed and direction of motion; The cloud morphological characteristics include the cloud volume and edge density of the target cloud; The cloud evolution characteristics include the fragmentation rate of the target cloud, which characterizes the rate of change of the total length of the edge of the target cloud.

5. The solar irradiance prediction method according to claim 3, characterized in that, The semantic segmentation network model is trained using the Dice loss function, conforming to the following formula: in, It represents the probability predicted by the model that the i-th pixel is a cloud. It is a truth value.

6. The solar irradiance prediction method according to claim 1 or 4, characterized in that, The training process of the regression model includes: The real-time horizontal projection distance of the target cloud body is determined based on the cloud body morphology and cloud body evolution characteristics of the target cloud body. The time delay of the target cloud is determined based on the target cloud's movement speed and real-time horizontal projection distance; The regression model is trained by using the sum of the time delay and the predicted duration as the model input and the predicted solar irradiance after the predicted duration as the model output.

7. The solar irradiance prediction method according to claim 1, characterized in that, After obtaining the predicted solar irradiance below the target UAV's flight path, the method further includes: Compare the predicted solar irradiance results with a preset safety threshold; If the predicted solar irradiance is lower than the safety threshold, a safety warning will be issued.

8. A solar irradiance prediction system, characterized in that, include: The data acquisition module is used to acquire cloud base image data of the target cloud body collected by the target drone; The semantic segmentation module is used to perform semantic segmentation on the cloud base image data after attitude correction to obtain a binarized cloud mask sequence. The feature extraction module is used to extract features from the binarized cloud mask sequence; The irradiance prediction module is used to input the extracted feature vectors into a pre-trained regression model to obtain the predicted solar irradiance below the flight path of the target UAV.

9. A solar irradiance prediction device, installed on a target UAV, characterized in that, include: The data processing module is used to execute the solar irradiance prediction method according to any one of claims 1-7; The image acquisition module includes at least one camera for continuously acquiring image sequences of the target cloud above the flight path of the target UAV; The pose perception module is used to acquire the speed and attitude of the target UAV. The wireless communication module is used for data communication with the ground station.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the solar irradiance prediction method as described in any one of claims 1-7.

Images