Dynamic cloud shadow trajectory prediction method and system based on machine learning algorithm

Abstract

The application relates to a dynamic cloud-shading trajectory prediction method and system based on a machine learning algorithm. The method comprises the following steps: collecting historical cloud layer movement data, meteorological data and real-time photosensitive panel output data of a target area, and performing a pretreatment operation; constructing a long short-term memory neural network model, inputting the pretreated data into the long short-term memory neural network model for training, and obtaining a cloud layer movement prediction model; collecting cloud layer image data and meteorological data in real time, inputting the data into the cloud layer movement prediction model for cloud layer movement trajectory prediction, and obtaining a cloud layer movement trajectory prediction result; and controlling a heliostat to perform dynamic angle adjustment according to the cloud layer movement trajectory prediction result, so that cloud layer movement tracking is realized. Through the LSTM neural network model, the cloud layer movement prediction model obtained through training can accurately mine the correlation rules between data, and the future cloud layer movement trajectory can be quickly predicted.

Description

Technical Field

[0001] This invention relates to the field of cloud trajectory prediction technology, and in particular to a method and system for predicting dynamic cloud occlusion trajectories based on machine learning algorithms. Background Technology

[0002] With the continuous development of solar photovoltaic power generation technology, large-scale centralized photovoltaic power plants and distributed photovoltaic power generation systems have been widely used. However, the output power of photovoltaic power generation systems is severely affected by cloud movement and shading, resulting in unstable power output and posing a significant challenge to grid dispatch and operation. To address the power fluctuation problem caused by cloud shading, existing technologies typically employ the installation of photosensitive sensors on heliostats and other equipment to indirectly predict the impact of clouds by monitoring changes in light intensity in different directions, and then adjusting the aiming direction of the heliostat based on the prediction results to attempt to maintain system efficiency. However, these methods have significant limitations: insufficient response speed to clouds, especially rapidly moving intermittent cloud shading, makes it difficult to adjust the heliostat attitude in a timely and accurate manner, leading to a significant decrease in efficiency during cloud shadow passage; at the data processing level, the lack of optimized screening and cleaning processes for cloud images and other data makes it impossible to effectively remove unqualified data, affecting the accuracy of subsequent analysis; in terms of model building, existing methods fail to fully utilize the dynamic correlation between historical cloud movement data and real-time photosensitive panel output, and also fail to deeply integrate meteorological data, making it difficult to establish accurate dynamic cloud shading trajectory prediction models.

[0003] Therefore, traditional methods of cloud prediction often suffer from poor quality cloud image data and low accuracy in cloud trajectory prediction because they cannot respond quickly to intermittent cloud cover and lack analysis of historical cloud movement data and real-time photosensitive panel output. Summary of the Invention

[0004] Therefore, in order to solve the above-mentioned technical problems, a method and system for predicting dynamic cloud occlusion trajectories based on machine learning algorithms are provided, which can improve the accuracy of cloud occlusion trajectory prediction.

[0005] A method for predicting dynamic cloud occlusion trajectories based on machine learning algorithms, the method comprising:

[0006] Historical cloud movement data, meteorological data, and real-time output data from the photosensitive panel are collected for the target area. The collected data is then preprocessed to obtain the preprocessed data.

[0007] A long short-term memory network model is constructed, and the preprocessed data is input into the long short-term memory neural network model for training. The weights and bias parameters of the long short-term memory network model are adjusted through the backpropagation algorithm to obtain a cloud movement prediction model.

[0008] Real-time cloud image data and meteorological data are collected and input into the cloud movement prediction model to predict the cloud movement trajectory, and the cloud movement trajectory prediction result is obtained.

[0009] The elevation angle of the heliostat is calculated based on the cloud motion trajectory prediction results, and the heliostat is controlled to dynamically adjust its angle according to the elevation angle to achieve cloud motion tracking.

[0010] In one embodiment, the collected data is preprocessed to obtain preprocessed data, including:

[0011] Remove data from the historical cloud movement data that have a resolution lower than a preset threshold, a missing area ratio exceeding a preset proportion, or an image blur greater than a preset value, to obtain filtered cloud movement data.

[0012] The filtered cloud movement data is cleaned, and a median filtering algorithm is used to remove noise and outliers to obtain cleaned data.

[0013] The cleaned data, meteorological data, and real-time photosensitive panel output data are normalized to obtain preprocessed data.

[0014] In one embodiment, constructing a long short-term memory neural network model includes:

[0015] A model architecture consisting of an input layer, at least one LSTM hidden layer, and an output layer is constructed. The input layer converts the input multi-source discrete temporal data into a temporal tensor. The LSTM hidden layer is set as a concatenated layer. The output layer uses a linear activation function.

[0016] The training parameters are configured for the model architecture, and the Adam optimizer is used to drive the parameter update. The mean squared error is selected as the loss function to construct the long short-term memory LSTM neural network model.

[0017] In one embodiment, the preprocessed data is input into the Long Short-Term Memory (LSTM) network model for training. The weights and bias parameters of the LTM network model are adjusted using a backpropagation algorithm to obtain a cloud movement prediction model, including:

[0018] The preprocessed historical cloud movement data, meteorological data, and real-time photosensitive panel output data are used as input features and input into the LSTM neural network model.

[0019] The LSTM neural network model is trained, and the backpropagation algorithm is used to adjust the model weights and bias parameters during the training process to obtain the trained model.

[0020] The trained model is validated using a test set. When the root mean square error of the trained model is lower than a preset threshold, the trained model is used as a cloud movement prediction model.

[0021] In one embodiment, cloud image data and meteorological data are acquired in real time and input into the cloud movement prediction model to predict the cloud movement trajectory, resulting in the cloud movement trajectory prediction result, including:

[0022] Real-time cloud image data and meteorological data of the target area are collected and converted into three-dimensional tensors and one-dimensional numerical sequences, respectively. The converted data are then normalized and data completion are performed to obtain preprocessed real-time data.

[0023] The preprocessed real-time data is input into the cloud mobility prediction model, the forward propagation process is invoked, the data is received through the input layer and passed to the LSTM hidden layer to extract temporal features, and then the prediction result is generated after integration and mapping through the output layer.

[0024] The prediction results are presented in a structured form, including prediction timestamps, trajectory data at each time point, and model inference confidence, and the cloud motion trajectory prediction results are output.

[0025] In one embodiment, the method further includes:

[0026] Extract the cloud motion parameters corresponding to the target time point from the cloud motion trajectory prediction results;

[0027] The installation position of the heliostat and the preset effective light-receiving elevation angle range are obtained. Based on the cloud movement parameters, it is predicted whether the cloud will block the solar radiation receiving path of the heliostat at each time point. The time points with obstruction are marked as the time points to be adjusted.

[0028] In one embodiment, calculating the elevation angle of the heliostat based on the cloud motion trajectory prediction result includes:

[0029] Obtain the solar altitude angle and solar azimuth angle corresponding to the time point to be adjusted;

[0030] The optimal illumination elevation angle of the heliostat under unobstructed conditions is calculated based on the solar elevation angle, and the obstruction influence coefficient is determined based on the cloud position coordinates in the cloud movement trajectory prediction results.

[0031] Determine the cloud velocity adjustment correction factor, and calculate the elevation angle of the heliostat based on the cloud velocity adjustment correction factor, the optimal light-gathering elevation angle, and the shading influence factor.

[0032] In one embodiment, controlling the heliostat to dynamically adjust its angle according to the elevation angle to achieve cloud motion tracking includes:

[0033] The elevation angles corresponding to each time point to be adjusted are converted into electrical signals according to the time axis sequence, and the electrical signals are sent to the heliostat's drive controller.

[0034] The drive controller controls the elevation angle adjustment mechanism to perform dynamic angle adjustment.

[0035] A dynamic cloud occlusion trajectory prediction system based on machine learning algorithms, the system comprising:

[0036] The data acquisition and preprocessing module is used to collect historical cloud movement data, meteorological data, and real-time photosensitive panel output data of the target area, and to perform preprocessing operations on the collected data to obtain preprocessed data.

[0037] The model building and training module is used to build a long short-term memory network model and input the preprocessed data into the long short-term memory neural network model for training. The weights and bias parameters of the long short-term memory network model are adjusted through the backpropagation algorithm to obtain the cloud movement prediction model.

[0038] The trajectory prediction module is used to collect cloud image data and meteorological data in real time and input them into the cloud movement prediction model to predict the cloud movement trajectory and obtain the cloud movement trajectory prediction result.

[0039] The heliostat adjustment module is used to calculate the elevation angle of the heliostat based on the cloud motion trajectory prediction result, and control the heliostat to dynamically adjust the angle according to the elevation angle to achieve cloud motion tracking.

[0040] The aforementioned dynamic cloud obscuration trajectory prediction method and system based on machine learning algorithms ensures data quality by collecting historical cloud movement data, meteorological data, and real-time photosensitive panel output data and performing preprocessing, thus providing high-quality samples for model training. Through an LSTM neural network model, the trained cloud movement prediction model can accurately uncover the correlation patterns between data points, enabling rapid prediction of future cloud movement trajectories. Furthermore, based on the real-time predicted cloud movement trajectory results, the optimal elevation angle of the heliostat is dynamically calculated, and the mechanical structure is driven to adjust in real time, achieving precise tracking of cloud movement. Attached Figure Description

[0041] Figure 1 This is an application environment diagram of a dynamic cloud occlusion trajectory prediction method based on machine learning algorithms in one embodiment;

[0042] Figure 2 This is a flowchart illustrating a dynamic cloud occlusion trajectory prediction method based on machine learning algorithms in one embodiment.

[0043] Figure 3This is a block diagram of a dynamic cloud occlusion trajectory prediction system based on machine learning algorithms in one embodiment;

[0044] Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0046] The dynamic cloud occlusion trajectory prediction method based on machine learning algorithms provided in this application can be applied to, for example... Figure 1 The application environment shown. For example... Figure 1 As shown, the application environment includes computer device 110. Computer device 110 can collect historical cloud movement data, meteorological data, and real-time photosensitive panel output data for the target area, and perform preprocessing operations on the collected data to obtain preprocessed data. Computer device 110 can construct a long short-term memory neural network model and input the preprocessed data into the long short-term memory neural network model for training. The weights and bias parameters of the long short-term memory network model are adjusted through backpropagation algorithm to obtain a cloud movement prediction model. Computer device 110 can collect cloud image data and meteorological data in real time and input them into the cloud movement prediction model to predict cloud movement trajectories, obtaining cloud movement trajectory prediction results. Computer device 110 can calculate the elevation angle of the heliostat based on the cloud movement trajectory prediction results and control the heliostat to dynamically adjust its angle according to the elevation angle, thereby achieving cloud movement tracking. The computer device 110 can be, but is not limited to, various personal computers, laptops, smartphones, robots, unmanned aerial vehicles, tablets, and other devices.

[0047] In one embodiment, such as Figure 2 As shown, a method for predicting dynamic cloud occlusion trajectories based on machine learning algorithms is provided, including the following steps:

[0048] Step 202: Collect historical cloud movement data, meteorological data, and real-time photosensitive panel output data of the target area, and perform preprocessing operations on the collected data to obtain preprocessed data.

[0049] The target area is the region where the solar concentrated solar power (CSP) plant is located. Computer equipment can collect historical satellite cloud imagery data of the target area, serving as the core foundational data for learning cloud movement patterns. The collection timeframe can be set to the past 1-3 years; the time resolution should be set to 15 minutes to 1 hour, ensuring the continuity of the time-series data while avoiding reduced training efficiency due to excessive data volume. The collected historical cloud movement data must include key information such as the spatial distribution, morphological changes, and coverage of clouds over the target area. The data format should be standardized as a standard image format to facilitate subsequent preprocessing operations.

[0050] Simultaneously, the computer equipment can collect historical and real-time meteorological data of the target area. The collected meteorological parameters include temperature, relative humidity, wind speed, and wind direction. The historical meteorological data and historical cloud movement data maintain the same time range and time resolution to ensure data time sequence matching. Real-time meteorological data is collected synchronously with subsequent real-time cloud image data. The meteorological data is collected through meteorological monitoring stations equipped with high-precision temperature sensors, humidity sensors, anemometers, and wind vanes to ensure the accuracy and stability of data collection.

[0051] By installing photosensitive panels on the heliostat, real-time radiation intensity data at different elevation angles can be collected. The photosensitive panels are evenly distributed along the edge of the heliostat's reflective surface, enabling real-time sensing of changes in solar radiation intensity. Their acquisition frequency is consistent with real-time meteorological data and real-time cloud image data. The collected data directly reflects the heliostat's current lighting status.

[0052] In one embodiment, the provided method for predicting dynamic cloud occlusion trajectories based on machine learning algorithms may further include a preprocessing operation, specifically including: removing data from historical cloud movement data that has a resolution lower than a preset threshold, a missing region ratio exceeding a preset proportion, or an image blur greater than a preset value, to obtain filtered cloud movement data; cleaning the filtered cloud movement data and using a median filtering algorithm to eliminate noise and outliers, to obtain cleaned data; and normalizing the cleaned data, meteorological data, and real-time photosensitive panel output data to obtain preprocessed data.

[0053] After data collection, the computer equipment can preprocess the collected data, including data filtering, data cleaning, and data standardization. Specifically, the computer equipment can filter the collected historical cloud movement data, removing unqualified cloud image data with a cloud image resolution lower than 500×500 pixels, containing more than 20% missing areas, or with an image blur greater than 0.8; the filtered cloud image data is then cleaned using methods such as median filtering to eliminate noise and outliers in the images; and the cleaned data is then standardized, normalizing the cloud image pixel values, meteorological data, and photosensitive panel output data to numerical ranges of 0-1, -1-1, and 0-1, respectively.

[0054] Step 204: Construct a long short-term memory neural network model and input the preprocessed data into the long short-term memory neural network model for training. Adjust the weights and bias parameters of the long short-term memory network model through the backpropagation algorithm to obtain the cloud movement prediction model.

[0055] Computer equipment can input preprocessed data into a Long Short-Term Memory (LSTM) neural network model for training, generating a cloud movement prediction model.

[0056] In one embodiment, a dynamic cloud occlusion trajectory prediction method based on machine learning algorithms may further include a model building process, specifically including: building a model architecture with an input layer, at least one LSTM hidden layer, and an output layer; the input layer converts the input multi-source discrete time-series data into a time-series tensor; the LSTM hidden layer is set as a concatenated layer; the output layer uses a linear activation function; training parameters are configured for the model architecture, and the Adam optimizer is used to drive parameter updates, with mean squared error selected as the loss function, to build a long short-term memory LSTM neural network model.

[0057] The constructed Long Short-Term Memory (LSTM) neural network model comprises an input layer, multiple LSTM hidden layers, and a fully connected output layer. Specifically, it is first necessary to clarify the core task and input / output of the LSTM neural network model. In this embodiment, the core task is to learn the correlation between multi-source time-series data and cloud movement trajectories, and to achieve time-series prediction of cloud position coordinates, movement speed, and movement direction for the next 1-4 hours. The input features are preprocessed historical / real-time cloud movement data sequences, real-time photosensitive panel output data sequences, and meteorological data sequences. The output result is a structured cloud trajectory prediction sequence with a time resolution of 15 minutes / time.

[0058] An architecture can be adopted, consisting of an input layer, multiple LSTM hidden layers, and a fully connected output layer. The input layer converts multi-source discrete temporal data into a temporal tensor that the model can recognize. The input dimension matches the total dimension of the multi-source features, and the time step is consistent with the data acquisition resolution. The LSTM hidden layer consists of three cascaded layers, each containing 128 neurons. The first two layers use a sequence return mode with the tanh activation function. The fully connected output layer uses a linear activation function, and the output dimension matches the structured dimension of the three prediction results: position, velocity, and orientation.

[0059] Next, the computer equipment can be configured with training hyperparameters and optimization strategies, setting the batch size to 64, the learning rate to 0.001, and the number of training epochs to 50-200. The Adam optimizer is used to drive parameter updates, and the mean squared error (MSE) is selected as the loss function to measure the difference between the predicted and actual values. The preprocessed data is then divided into training, validation, and test sets in a 7:2:1 ratio. The training set data is batch-input into the model to complete forward propagation, generate prediction results, and calculate the loss value. The loss value is then backpropagated using the backpropagation algorithm to adaptively adjust the weights and bias parameters of each layer of the model. During training, the validation set loss value is monitored; if it does not decrease for 10 consecutive epochs, an early stopping strategy is triggered.

[0060] Finally, the root mean square error (RMSE) can be used as the core validation metric. The test set data is input into the trained model, and the RMSE between the predicted result and the true value is calculated. When the root mean square error (RMSE) is less than 0.15, the final cloud movement prediction model is generated.

[0061] In one embodiment, a dynamic cloud obscuring trajectory prediction method based on machine learning algorithms may further include a process of training a cloud movement prediction model. The specific process includes: inputting preprocessed historical cloud movement data, meteorological data, and real-time photosensitive panel output data as input features into an LSTM neural network model; training the LSTM neural network model and adjusting the model weights and bias parameters using a backpropagation algorithm during training to obtain the trained model; validating the trained model using a test set, and using the trained model as the cloud movement prediction model when the root mean square error of the trained model is lower than a preset threshold.

[0062] In other words, computer equipment can use preprocessed historical cloud movement data, meteorological data, and real-time photosensitive panel output data as input features to train an LSTM neural network model. During training, the weights and bias parameters of the LSTM neural network model are continuously adjusted through the backpropagation algorithm, enabling the model to learn the cloud movement patterns. The trained LSTM neural network model is then validated by calculating metrics such as the root mean square error (RMSE) on the test set. When the model performance meets the requirements, the final cloud movement prediction model is obtained.

[0063] Step 206: Collect cloud image data and meteorological data in real time and input them into the cloud movement prediction model to predict the cloud movement trajectory and obtain the cloud movement trajectory prediction result.

[0064] Computer equipment can use generated cloud movement prediction models to predict real-time cloud data and obtain cloud trajectory prediction results. Specifically, the computer equipment can input real-time collected cloud image data and meteorological data into a trained LSTM neural network model, and the model will output the cloud trajectory prediction results for the next 1-4 hours, including information such as cloud position, speed, and direction.

[0065] In one embodiment, a dynamic cloud obscuration trajectory prediction method based on machine learning algorithms may further include a process for predicting cloud movement trajectories. The specific process includes: real-time acquisition of cloud image data and meteorological data of the target area, converting them into three-dimensional tensors and one-dimensional numerical sequences respectively; normalizing and completing the converted data to obtain preprocessed real-time data; inputting the preprocessed real-time data into a cloud movement prediction model; invoking the forward propagation process; receiving data through the input layer and transmitting it to the LSTM hidden layer to extract temporal features; then integrating and mapping the data through the output layer to generate prediction results; presenting the prediction results in a structured form, including prediction timestamps, trajectory data at each time point, and model inference confidence; and outputting the cloud movement trajectory prediction results.

[0066] Computer equipment can synchronously collect cloud image data of the target area through satellite cloud image receiving units and ground cloud image monitoring cameras, and collect real-time meteorological data through temperature sensors, humidity sensors, anemometers, and wind vanes at meteorological monitoring stations. The collected cloud image data and real-time meteorological data can be synchronized using the NTP time synchronization protocol to maintain consistent timestamps, with a collection frequency of once every 15 minutes. The cloud image data resolution is no less than 500×500 pixels, and the meteorological data includes real-time temperature, relative humidity, wind speed, and wind direction parameters.

[0067] Next, the cloud image data can be converted into a three-dimensional tensor in H×W×C format, and the meteorological data can be arranged into a one-dimensional numerical sequence in the order of temperature-humidity-wind speed-wind direction. Through the normalization parameters used in the model training phase, the pixel values ​​of the cloud images are normalized to the range of 0-1, and the meteorological data is normalized to the range of -1-1. If the cloud image resolution is lower than 500×500 pixels, the missing area accounts for more than 20%, or the blurriness is greater than 0.8, the computer equipment can use interpolation of cloud images from adjacent time points to supplement the data. If meteorological parameters are missing or exceed a reasonable physical range, valid data from the previous acquisition period is used to complete the data.

[0068] By preloading a cloud mobility prediction model with fixed parameters, the preprocessed real-time data is input into the model once, and the forward propagation process is called. The data is received through the input layer and passed to the LSTM hidden layer. The LSTM hidden layer extracts the temporal features in the real-time data, and then the prediction results are generated after being integrated and mapped by the fully connected output layer.

[0069] The system outputs structured trajectory data in sets of 15 minutes every 1-4 hours, including cloud position coordinates, cloud movement speed, and cloud movement direction with the power plant center point as the origin. In this embodiment, the cloud trajectory prediction results are presented in JSON format, including the prediction timestamp, trajectory data at each time point, and model inference confidence, where the confidence is calculated based on the similarity between real-time data and training data.

[0070] Step 208: Calculate the elevation angle of the heliostat based on the cloud motion trajectory prediction results, and control the heliostat to dynamically adjust its angle according to the elevation angle to achieve cloud motion tracking.

[0071] Computer equipment can dynamically adjust the elevation angle of heliostats based on cloud movement trajectory predictions, improving the response speed to intermittent shading. Specifically, the computer equipment can calculate the required elevation angle for the heliostats based on cloud movement trajectory predictions, enabling them to track cloud movement in a timely manner; controlling the heliostats to adjust according to the calculated elevation angle further improves the response speed to intermittent shading, thereby increasing the overall efficiency of the concentrated solar power (CSP) system.

[0072] In one embodiment, a dynamic cloud shading trajectory prediction method based on machine learning algorithms may further include the process of determining the time point to be adjusted. Specifically, this includes: extracting the cloud motion parameters corresponding to the target time point from the cloud motion trajectory prediction results; obtaining the heliostat installation position and the preset effective light-receiving elevation angle range; and predicting whether the cloud will block the solar radiation receiving path of the heliostat at each time point based on the cloud motion parameters, and marking the time points with blockage as the time points to be adjusted.

[0073] The computer equipment can extract the core parameters corresponding to every 15 minutes in the cloud movement trajectory prediction results for the next 1-4 hours, including the cloud position coordinates with the center point of the power station as the origin, the cloud movement speed and the cloud movement direction; combined with the heliostat installation position and the preset effective light-receiving elevation angle range, it can predict whether the cloud will block the solar radiation receiving path of the heliostat at each time point and mark the time points to be adjusted.

[0074] In one embodiment, a method for predicting dynamic cloud shading trajectories based on machine learning algorithms may further include the process of calculating the elevation angle of a heliostat. The specific process includes: obtaining the solar altitude angle and solar azimuth angle corresponding to the time point to be adjusted; calculating the optimal illumination elevation angle of the heliostat when there is no shading based on the solar altitude angle, and determining the shading influence coefficient based on the cloud position coordinates in the cloud movement trajectory prediction result; determining the cloud velocity adjustment correction coefficient, and calculating the elevation angle of the heliostat based on the cloud velocity adjustment correction coefficient, the optimal illumination elevation angle, and the shading influence coefficient.

[0075] The computer equipment can obtain the solar altitude angle and solar azimuth angle at the time point to be adjusted, and call up hardware parameters such as the maximum / minimum elevation angle limit of the heliostat and the upper limit of the adjustment angular velocity. First, it calculates the optimal illumination elevation angle when there is no obstruction based on the solar altitude angle, then determines the obstruction influence coefficient according to the cloud position coordinates, and calculates the target elevation angle in combination with the cloud velocity adjustment correction coefficient. The target elevation angle must be within the elevation angle range allowed by the heliostat hardware. If it is a centralized power station with multiple heliostats, the target elevation angle of each heliostat is calculated separately according to the distribution of the heliostat array, and the adjustment time difference between adjacent heliostats is set to avoid grid load fluctuations.

[0076] In one embodiment, a method for predicting dynamic cloud obscuration trajectories based on machine learning algorithms may further include a process of adjusting the angle of a heliostat. The specific process includes: converting the elevation angle corresponding to each time point to be adjusted into an electrical signal in chronological order, and sending the electrical signal to the drive controller of the heliostat; the drive controller controls the elevation angle adjustment mechanism to perform dynamic angle adjustment.

[0077] The computer equipment can convert the target elevation angle at each time point into electrical signal commands in chronological order, and send them to the heliostat drive controller via industrial bus or 5G communication module. The drive controller controls the elevation angle adjustment mechanism. When the difference between the current elevation angle and the target elevation angle is greater than 10°, it gradually approaches the target elevation angle at the maximum angular velocity allowed by the hardware. When the difference is less than or equal to 10°, it slowly adjusts to achieve a smooth transition. The angle sensor on the heliostat collects the actual elevation angle in real time and feeds it back to the drive controller to form a closed-loop control.

[0078] In this embodiment, the computer device can also collect adjusted radiation intensity data through the photosensitive panel, and combine it with real-time cloud image data to verify the lighting efficiency and occlusion status; if the radiation intensity does not reach the preset threshold or there is still occlusion, the cloud movement prediction model is called again to obtain the latest trajectory prediction results within 30 minutes, the target elevation angle is recalculated and a secondary adjustment is triggered; the adjustment deviation data is recorded and fed back to the model optimization module for iterative update of LSTM neural network model parameters.

[0079] It should be understood that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart above may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0080] In one embodiment, such as Figure 3 As shown, a dynamic cloud occlusion trajectory prediction system based on machine learning algorithms is provided, including: a data acquisition and preprocessing module 310, a model building and training module 320, a trajectory prediction module 330, and a heliostat adjustment module 340, wherein:

[0081] The data acquisition and preprocessing module 310 is used to acquire historical cloud movement data, meteorological data, and real-time photosensitive panel output data of the target area, and to perform preprocessing operations on the acquired data to obtain preprocessed data.

[0082] The model building and training module 320 is used to build a long short-term memory neural network model and input the preprocessed data into the long short-term memory neural network model for training. The weights and bias parameters of the long short-term memory network model are adjusted through the backpropagation algorithm to obtain the cloud movement prediction model.

[0083] The trajectory prediction module 330 is used to collect cloud image data and meteorological data in real time and input them into the cloud movement prediction model to predict the cloud movement trajectory and obtain the cloud movement trajectory prediction result.

[0084] The heliostat adjustment module 340 is used to calculate the elevation angle of the heliostat based on the cloud motion trajectory prediction results, and control the heliostat to dynamically adjust the angle according to the elevation angle to achieve cloud motion tracking.

[0085] In one embodiment, the data acquisition and preprocessing module 310 is further used to remove data from historical cloud movement data that has a resolution lower than a preset threshold, a missing area ratio exceeding a preset proportion, or an image blur greater than a preset value, to obtain filtered cloud movement data; to perform data cleaning on the filtered cloud movement data and to use a median filtering algorithm to eliminate noise and outliers, to obtain cleaned data; and to perform normalization processing on the cleaned data, meteorological data, and real-time photosensitive panel output data to obtain preprocessed data.

[0086] In one embodiment, the model building and training module 320 is also used to build a model architecture consisting of an input layer, at least one LSTM hidden layer, and an output layer; the input layer converts the input multi-source discrete time series data into a time series tensor; the LSTM hidden layer is set as a concatenated layer; the output layer uses a linear activation function; training parameters are configured for the model architecture, and the Adam optimizer is used to drive parameter updates, with mean squared error selected as the loss function to build a long short-term memory LSTM neural network model.

[0087] In one embodiment, the model building and training module 320 is further used to input preprocessed historical cloud movement data, meteorological data, and real-time photosensitive panel output data as input features into the LSTM neural network model; train the LSTM neural network model, and adjust the model weights and bias parameters using the backpropagation algorithm during the training process to obtain the trained model; use the test set to verify the trained model, and when the root mean square error of the trained model is lower than a preset threshold, the trained model is used as the cloud movement prediction model.

[0088] In one embodiment, the trajectory prediction module 330 is further configured to collect cloud image data and meteorological data of the target area in real time, convert them into three-dimensional tensors and one-dimensional numerical sequences respectively, normalize and complete the converted data to obtain preprocessed real-time data; input the preprocessed real-time data into the cloud movement prediction model, call the forward propagation process, receive data through the input layer and pass it to the LSTM hidden layer to extract temporal features, and then generate prediction results after integration and mapping through the output layer; present the prediction results in a structured form, including prediction timestamps, trajectory data at each time point and model inference confidence, and output the cloud movement trajectory prediction results.

[0089] In one embodiment, the heliostat adjustment module 340 is further used to extract the cloud motion parameters corresponding to the target time point in the cloud motion trajectory prediction result; obtain the heliostat installation position and the preset effective light-receiving elevation angle range; and predict whether the cloud will block the solar radiation receiving path of the heliostat at each time point according to the cloud motion parameters, and mark the time points with obstruction as the time points to be adjusted.

[0090] In one embodiment, the heliostat adjustment module 340 is further configured to obtain the solar altitude angle and solar azimuth angle corresponding to the time point to be adjusted; calculate the optimal illumination elevation angle of the heliostat when there is no obstruction based on the solar altitude angle, and determine the obstruction influence coefficient according to the cloud position coordinates in the cloud movement trajectory prediction result; determine the cloud speed adjustment correction coefficient, and calculate the elevation angle of the heliostat according to the cloud speed adjustment correction coefficient, the optimal illumination elevation angle, and the obstruction influence coefficient.

[0091] In one embodiment, the heliostat adjustment module 340 is further configured to convert the elevation angle corresponding to each time point to be adjusted into an electrical signal in sequence according to the time axis, and send the electrical signal to the heliostat drive controller; the drive controller controls the elevation angle adjustment mechanism to perform dynamic angle adjustment.

[0092] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements a dynamic cloud occlusion trajectory prediction method based on machine learning algorithms. The display screen can be an LCD screen or an e-ink display screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0093] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0094] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a dynamic cloud occlusion trajectory prediction method based on a machine learning algorithm.

[0095] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of a dynamic cloud occlusion trajectory prediction method based on a machine learning algorithm.

[0096] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0097] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting dynamic cloud occlusion trajectories based on machine learning algorithms, characterized in that, The method includes: Historical cloud movement data, meteorological data, and real-time output data from the photosensitive panel are collected for the target area. The collected data is then preprocessed to obtain the preprocessed data. A long short-term memory neural network model is constructed, and the preprocessed data is input into the long short-term memory neural network model for training. The weights and bias parameters of the long short-term memory network model are adjusted by the backpropagation algorithm to obtain a cloud movement prediction model. Real-time cloud image data and meteorological data are collected and input into the cloud movement prediction model to predict the cloud movement trajectory, and the cloud movement trajectory prediction result is obtained. The elevation angle of the heliostat is calculated based on the cloud motion trajectory prediction results, and the heliostat is controlled to dynamically adjust its angle according to the elevation angle to achieve cloud motion tracking.

2. The dynamic cloud occlusion trajectory prediction method based on machine learning algorithm according to claim 1, characterized in that, The collected data is preprocessed to obtain preprocessed data, including: Remove data from the historical cloud movement data that have a resolution lower than a preset threshold, a missing area ratio exceeding a preset proportion, or an image blur greater than a preset value, to obtain filtered cloud movement data. The filtered cloud movement data is cleaned, and a median filtering algorithm is used to remove noise and outliers to obtain cleaned data. The cleaned data, meteorological data, and real-time photosensitive panel output data are normalized to obtain preprocessed data.

3. The dynamic cloud occlusion trajectory prediction method based on machine learning algorithm according to claim 1, characterized in that, Constructing a long short-term memory neural network model includes: A model architecture consisting of an input layer, at least one LSTM hidden layer, and an output layer is constructed. The input layer converts the input multi-source discrete temporal data into a temporal tensor. The LSTM hidden layer is set as a concatenated layer. The output layer uses a linear activation function. The training parameters are configured for the model architecture, and the Adam optimizer is used to drive the parameter update. The mean squared error is selected as the loss function to construct the long short-term memory LSTM neural network model.

4. The dynamic cloud occlusion trajectory prediction method based on machine learning algorithm according to claim 3, characterized in that, The preprocessed data is input into the Long Short-Term Memory (LSTM) network model for training. The weights and bias parameters of the LSM network model are adjusted using the backpropagation algorithm to obtain a cloud movement prediction model, including: The preprocessed historical cloud movement data, meteorological data, and real-time photosensitive panel output data are used as input features and input into the LSTM neural network model. The LSTM neural network model is trained, and the backpropagation algorithm is used to adjust the model weights and bias parameters during the training process to obtain the trained model. The trained model is validated using a test set. When the root mean square error of the trained model is lower than a preset threshold, the trained model is used as a cloud movement prediction model.

5. The dynamic cloud occlusion trajectory prediction method based on machine learning algorithm according to claim 1, characterized in that, Real-time cloud image data and meteorological data are collected and input into the cloud movement prediction model to predict cloud movement trajectories, resulting in cloud movement trajectory prediction results, including: Real-time cloud image data and meteorological data of the target area are collected and converted into three-dimensional tensors and one-dimensional numerical sequences, respectively. The converted data are then normalized and data completion are performed to obtain preprocessed real-time data. The preprocessed real-time data is input into the cloud mobility prediction model, the forward propagation process is invoked, the data is received through the input layer and passed to the LSTM hidden layer to extract temporal features, and then the prediction result is generated after integration and mapping through the output layer. The prediction results are presented in a structured form, including prediction timestamps, trajectory data at each time point, and model inference confidence, and the cloud motion trajectory prediction results are output.

6. The dynamic cloud occlusion trajectory prediction method based on machine learning algorithm according to claim 1, characterized in that, The method further includes: Extract the cloud motion parameters corresponding to the target time point from the cloud motion trajectory prediction results; The installation position of the heliostat and the preset effective light-receiving elevation angle range are obtained. Based on the cloud movement parameters, it is predicted whether the cloud will block the solar radiation receiving path of the heliostat at each time point. The time points with obstruction are marked as the time points to be adjusted.

7. The dynamic cloud occlusion trajectory prediction method based on machine learning algorithm according to claim 6, characterized in that, The elevation angle of the heliostat is calculated based on the cloud motion trajectory prediction results, including: Obtain the solar altitude angle and solar azimuth angle corresponding to the time point to be adjusted; The optimal illumination elevation angle of the heliostat under unobstructed conditions is calculated based on the solar elevation angle, and the obstruction influence coefficient is determined based on the cloud position coordinates in the cloud movement trajectory prediction results. Determine the cloud velocity adjustment correction factor, and calculate the elevation angle of the heliostat based on the cloud velocity adjustment correction factor, the optimal light-gathering elevation angle, and the shading influence factor.

8. The dynamic cloud occlusion trajectory prediction method based on machine learning algorithm according to claim 7, characterized in that, Controlling the heliostat to dynamically adjust its angle according to the elevation angle to achieve cloud movement tracking includes: The elevation angles corresponding to each time point to be adjusted are converted into electrical signals according to the timeline sequence, and the electrical signals are sent to the heliostat's drive controller. The drive controller controls the elevation angle adjustment mechanism to perform dynamic angle adjustment.

9. A dynamic cloud occlusion trajectory prediction system based on machine learning algorithms, characterized in that, The system includes: The data acquisition and preprocessing module is used to collect historical cloud movement data, meteorological data, and real-time photosensitive panel output data of the target area, and to perform preprocessing operations on the collected data to obtain preprocessed data. The model building and training module is used to build a long short-term memory neural network model, and input the preprocessed data into the long short-term memory neural network model for training. The weights and bias parameters of the long short-term memory network model are adjusted through the backpropagation algorithm to obtain a cloud movement prediction model. The trajectory prediction module is used to collect cloud image data and meteorological data in real time and input them into the cloud movement prediction model to predict the cloud movement trajectory and obtain the cloud movement trajectory prediction result. The heliostat adjustment module is used to calculate the elevation angle of the heliostat based on the cloud motion trajectory prediction result, and control the heliostat to dynamically adjust the angle according to the elevation angle to achieve cloud motion tracking.

Images