Biaxial light tracking and off-grid control system and method for photovoltaic power generation

Abstract

The present disclosure relates to a photovoltaic power generation dual-axis light tracking and off-grid control system and method, comprising: acquiring light data, weather forecast, satellite cloud image and environmental parameters, forming multi-modal photovoltaic data and constructing a light prediction model based on a machine learning method to predict the target solar irradiance and the light tracking angle; the photovoltaic platform performs overall rotation according to the predicted angle, and combines local fine tuning of each independent photovoltaic unit to realize a composite light tracking action sequence, and the multiple photovoltaic platforms share the prediction data through wireless communication to form a leading-following mechanism, when the leading photovoltaic platform detects that the light changes more than the threshold value, a correction instruction is generated and sent to other photovoltaic platforms, so that the light tracking action is corrected in advance. The present application can predict the solar irradiance and the optimal light tracking angle in advance and avoid the delay and misoperation caused by the independent response of each photovoltaic platform to the light change through group coordination.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation technology, and more specifically to a photovoltaic power generation dual-axis tracking and off-grid control system and method. Background Technology

[0002] Photovoltaic power generation, as a clean and renewable energy utilization method, has been widely used in distributed power generation and large-scale power plants in recent years. Since the output power of photovoltaic modules is closely related to the intensity of incident sunlight, the incident angle of solar energy has a decisive impact on the efficiency of photovoltaic power generation. To improve the overall power generation efficiency of photovoltaic systems, single-axis tracking and dual-axis tracking technologies have been proposed. Among them, the dual-axis tracking system can simultaneously adjust the azimuth and elevation angles to make the photovoltaic modules as perpendicular to the sunlight as possible, thereby achieving higher energy capture efficiency.

[0003] Currently, existing photovoltaic (PV) tracking platforms typically employ fixed astronomical algorithms or single illumination sensors for angle adjustment. While these methods can achieve good tracking results under clear weather conditions, their reliance on solar position formulas or single illumination sensor feedback makes them ill-suited for handling variable weather conditions, particularly light fluctuations caused by rapidly changing cloud cover or deteriorating air quality. Furthermore, in existing large-scale PV power plants, multiple PV platforms operate independently, lacking information sharing and coordinated control. When a pilot platform is shaded or experiences sudden changes in illumination, subsequent platforms continue to operate blindly, resulting in additional energy waste and a decrease in overall power generation efficiency. Summary of the Invention

[0004] This application provides a photovoltaic power generation dual-axis tracking and off-grid control system and method, which can achieve high-efficiency, low-energy consumption and group-coordinated photovoltaic tracking control.

[0005] In a first aspect, this application provides a method for dual-axis tracking and off-grid control of photovoltaic power generation, including a photovoltaic platform equipped with multiple independently adjustable photovoltaic power generation units. The method includes: acquiring irradiance data, weather forecast data, satellite cloud image data, and environmental parameters to form multimodal photovoltaic data; constructing an irradiance prediction model based on the multimodal photovoltaic data using machine learning methods, and outputting a target solar irradiance intensity and a target tracking angle; controlling the photovoltaic platform to rotate as a whole and making local independent adjustments to the multiple independently adjustable photovoltaic power generation units according to the target tracking angle, thereby realizing a composite tracking action sequence; and sharing the prediction data among the multiple photovoltaic platforms via wireless communication. When a lead photovoltaic platform detects that the irradiance change exceeds a preset threshold, it generates a correction command and sends it to other photovoltaic platforms, enabling the other photovoltaic platforms to correct the tracking action sequence in advance according to the received correction command and the frequency of irradiance change.

[0006] In one alternative of the first aspect, the construction of the illumination prediction model includes: modeling the time series of weather forecast data and satellite cloud image data using a linear regression method or a long short-term memory network to predict the illumination trend within a preset time range; inputting actual illumination data and environmental parameters into a correction layer to dynamically correct the illumination trend prediction results and generate a corrected illumination trend prediction value; and calculating the target solar irradiance and target tracking angle by combining the geographical location and time information of the photovoltaic platform.

[0007] In one optional embodiment of the first aspect, the implementation of the composite light-tracking action sequence includes: comparing the target light-tracking angle with the current posture of the photovoltaic platform and determining whether there is a deviation based on the comparison result; if so, calculating the overall adjustment compensation amount of the photovoltaic platform and controlling the overall rotation of the photovoltaic platform based on the overall adjustment compensation amount; calculating the independent compensation amount of each photovoltaic power generation unit based on the local illumination data of the location of each photovoltaic power generation unit of the photovoltaic platform and the target light-tracking angle, and controlling the corresponding photovoltaic power generation unit to adjust independently based on the independent compensation amount.

[0008] In one optional embodiment of the first aspect, the advance correction of the light-tracking action sequence includes: multiple photovoltaic platforms sharing illumination data, satellite cloud image data, and environmental parameters via wireless communication; selecting a photovoltaic platform located at a leading illumination position as a leading photovoltaic platform; when the leading photovoltaic platform detects that the illumination change exceeds a preset threshold, predicting the target light-tracking angle based on the illumination data, satellite cloud image data, and environmental parameters at the current location, and generating a correction command based on the target light-tracking angle and sending it to other photovoltaic platforms; and other photovoltaic platforms correcting the light-tracking action sequence in advance based on the received correction command and the frequency of illumination change.

[0009] In one alternative of the first aspect, when correcting the tracking action sequence in advance, the tracking action sequence of multiple photovoltaic platforms is dynamically optimized using an ant colony algorithm or a particle swarm optimization algorithm.

[0010] In one alternative of the first aspect, an off-grid energy storage battery is also included, and the method further includes: real-time detection of the state of charge of the off-grid energy storage battery; and dynamic adjustment of the solar tracking strategy according to the state of charge of the energy storage battery. The dynamic adjustment of the solar tracking strategy includes: entering a low-action energy-saving mode when the energy storage is close to full charge; and enhancing solar tracking to maximize power generation when the energy storage is insufficient.

[0011] In one alternative embodiment of the first aspect, when dynamically adjusting the tracking strategy based on the state of charge (SBC) of the energy storage battery, the method further includes: determining the SBC of each off-grid energy storage battery; if the SBC of the off-grid energy storage battery is greater than a first threshold, controlling the corresponding photovoltaic platform of the off-grid energy storage battery to enter a low-action energy-saving mode to reduce the number of angle adjustments; if the SBC of the off-grid energy storage battery is less than a second threshold, controlling the corresponding photovoltaic platform of the off-grid energy storage battery to enter a high-efficiency power generation mode to increase the angle adjustment frequency; if the SBC of the off-grid energy storage battery is between the first and second thresholds, controlling the corresponding photovoltaic platform of the off-grid energy storage battery to maintain a conventional tracking mode to balance and adjust the composite tracking action sequence according to the target tracking angle output by the prediction model.

[0012] In one alternative embodiment of the first aspect, when the photovoltaic platform executes the tracking action sequence, the method further includes: determining whether the photovoltaic sub-unit is faulty by comparing the power output of the photovoltaic power generation unit with the expected value of the prediction model; simultaneously calibrating the data collected by the sensor element to determine whether the data drifts or exceeds the safety threshold; if so, determining the photovoltaic platform corresponding to the fault or sensor element drift as an abnormal photovoltaic platform, and automatically correcting the tracking angle of the abnormal photovoltaic platform, or switching the abnormal photovoltaic platform to a redundant operation mode.

[0013] In one alternative embodiment of the first aspect, a photovoltaic cleaning element is included. When the photovoltaic platform performs a sequence of tracking actions, the method further includes: adjusting the photovoltaic platform from a normal power generation posture to a preset cleaning posture every first preset time interval, with at least a portion of the surface of the photovoltaic platform facing the cleaning area; controlling the photovoltaic cleaning element to enter a cleaning state for a second preset time interval, cleaning contaminants from the surface of the photovoltaic platform, and after cleaning is completed, adjusting the photovoltaic platform from the preset cleaning posture back to the normal power generation posture.

[0014] Secondly, this application provides a photovoltaic power generation dual-axis tracking and off-grid control system using the method according to any one of the first aspects, comprising: a photovoltaic platform including multiple independently adjustable photovoltaic power generation units and a first angle adjustment element, each photovoltaic power generation unit being provided with at least one second angle adjustment element; an off-grid energy storage battery connected to at least one photovoltaic power generation unit for storing energy collected by the connected photovoltaic power generation unit; sensor elements respectively disposed in the photovoltaic power generation unit and the photovoltaic power generation area for acquiring light intensity and environmental parameters including light intensity, temperature, humidity, wind speed, and air quality index of the target area; a photovoltaic cleaning element disposed in the photovoltaic power generation unit for cleaning the surface of the photovoltaic power generation unit using at least one method of vibration, brushing, air blowing, and electromagnetic tapping; and a control device connected to and controlling the photovoltaic platform, the off-grid energy storage battery, the sensor elements, and the photovoltaic cleaning elements to execute the method.

[0015] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0016] The accompanying drawings, which are incorporated herein and form part of this specification, illustrate one or more embodiments of the present application and, together with the description, serve to explain the principles of the present application and to enable those skilled in the art to make and use the present application.

[0017] Figure 1 This is a schematic diagram of the connection of an exemplary photovoltaic power generation dual-axis tracking and off-grid control system according to some embodiments of this application.

[0018] Figure 2 This is a schematic diagram of a regional tracking photovoltaic platform according to some embodiments of this application.

[0019] Figure 3 This is a schematic diagram of the structure of an exemplary photovoltaic platform according to some embodiments of this application.

[0020] Figure 4 This is a first light-tracking schematic diagram of an exemplary photovoltaic platform according to some embodiments of this application.

[0021] Figure 5 This is a second schematic diagram of an exemplary photovoltaic platform according to some embodiments of this application.

[0022] Figure 6 This is a schematic flowchart of an exemplary photovoltaic power generation dual-axis tracking and off-grid control method according to some embodiments of this application.

[0023] Figure 7 This is a flowchart illustrating an exemplary method for constructing an illumination prediction model according to some embodiments of this application.

[0024] Figure 8 This is a schematic flowchart illustrating an exemplary implementation of a composite light-tracking action sequence method according to some embodiments of this application.

[0025] Figure 9 This is a flowchart illustrating an exemplary multi-photovoltaic platform collaborative method according to some embodiments of this application.

[0026] Figure 10 This is a flowchart illustrating an exemplary method for dynamically adjusting a light-tracking strategy according to some embodiments of this application.

[0027] Figure 11 This is a flowchart illustrating an exemplary exception handling method according to some embodiments of this application.

[0028] Figure 12 This is a schematic flowchart of an exemplary periodic photovoltaic cleaning method according to some embodiments of this application.

[0029] Figure 13 This is a schematic diagram of the connection of an exemplary electronic device according to some embodiments of this application.

[0030] Explanation of reference numerals in the attached figures: 10. Photovoltaic power generation dual-axis tracking and off-grid control system; 11. Photovoltaic platform; 12. Off-grid energy storage battery; 13. Sensor element; 14. Control device; 15. Photovoltaic cleaning element; 11A. Photovoltaic power generation electronic unit; 11B. First angle adjustment element; 11C. Second angle adjustment element; 20. Photovoltaic power generation dual-axis tracking and off-grid control method; 30. Electronic equipment; 31. Memory; 32. Processor; 33. Communication interface. Detailed Implementation

[0031] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments may be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, the description of these embodiments is intended to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to provide a deeper understanding of embodiments of this application.

[0032] Currently, the existing dual-axis tracking platform is a device that can adjust the orientation of photovoltaic modules simultaneously in two degrees of freedom. Its main purpose is to keep the photovoltaic modules as perpendicular to the sunlight as possible under all weather conditions, thereby increasing the photovoltaic power generation. The two degrees of freedom mentioned above usually include two rotational degrees of freedom: azimuth angle (horizontal rotation) and pitch angle (vertical pitch).

[0033] Existing systems typically detect the sun's position using light sensors or calculate it based on astronomical algorithms, and then drive motors to adjust the position of photovoltaic modules in azimuth and / or pitch angles so that the photovoltaic modules are aligned with the sun in real time. Theoretically, dual-axis tracking platforms can improve power generation efficiency by 25%–40% compared to fixed photovoltaic modules, and are especially suitable for scenarios that require maximizing the power generation of a single unit, such as off-grid power supply and distributed independent power stations.

[0034] Currently, most existing dual-axis solar tracking platforms adopt an integral rotating support structure, where the entire photovoltaic platform is driven by a single large support rod and rotating joint to rotate the entire photovoltaic module. However, integral rotation involves a large mass of the module, resulting in a heavy load on the motor. Frequent adjustments lead to increased energy consumption, and in some applications, this even offsets the power generation gain brought by solar tracking. Furthermore, long-term, large-amplitude movements cause mechanical wear on the rotating joints, motors, and support structures, leading to increased maintenance costs. In addition, most existing photovoltaic solar tracking systems rely on fixed astronomical algorithms or a single light sensor for angle adjustment. While these methods can achieve good tracking results under clear weather conditions, in practical applications, existing solar tracking control mostly relies on solar position formulas or single-point sensor feedback, making it difficult to effectively cope with changing weather conditions, especially light fluctuations caused by rapid changes in cloud cover or deteriorating air quality (such as smog or dust storms). Moreover, because integral rotation lacks localized differential adjustment capabilities, it cannot fine-tune for light differences in different zones, thus limiting solar tracking efficiency.

[0035] Therefore, in order to address the shortcomings of existing dual-axis tracking platforms, reference was made to... Figures 1 to 5 As shown, Figure 1 This document illustrates a schematic diagram of the connection of an exemplary photovoltaic power generation dual-axis tracking and off-grid control system according to some embodiments of this application. Figure 2 A schematic diagram of a regional solar tracking system for an exemplary photovoltaic platform according to some embodiments of this application is shown. Figure 3 A schematic diagram of the structure of an exemplary photovoltaic platform according to some embodiments of this application is shown. Figure 4 A first schematic diagram of an exemplary photovoltaic platform according to some embodiments of this application is shown. Figure 5 This application illustrates a second schematic diagram of an exemplary photovoltaic platform according to some embodiments of the present application. The application relates to a photovoltaic power generation dual-axis tracking and off-grid control system 10, which includes at least a photovoltaic platform 11, an off-grid energy storage battery 12, sensor elements 13, and a control device 14.

[0036] The photovoltaic platform 11 includes multiple independently adjustable photovoltaic power generation units 11A and a first angle adjustment element 11B. Each photovoltaic power generation unit 10 is mounted on the photovoltaic platform 11 via a support bracket, and each photovoltaic power generation unit 11A is independently provided with at least one second angle adjustment element 11C.

[0037] The first angle adjustment element 11B is preferably driven by a stepper motor or a servo motor and is used to drive the overall rotation adjustment of the photovoltaic platform 11; the second angle adjustment element 11C is preferably driven by a stepper motor or a servo motor and is connected to the azimuth rotation axis and the pitch rotation axis respectively, thereby realizing dual-axis rotation adjustment and enabling the connected photovoltaic power generation unit 11A to independently adjust its attitude in two degrees of freedom, horizontal and vertical.

[0038] In actual implementation, the photovoltaic platform 11 can rotate as a whole, and can also perform local independent angle adjustments for different photovoltaic power generation units 11A, thereby supporting composite light tracking control.

[0039] The off-grid energy storage battery 12 is electrically connected to at least one photovoltaic power generation unit 11A for storing the energy collected by the photovoltaic power generation unit 11A.

[0040] In actual implementation, the off-grid energy storage battery 12 can be a lithium battery pack, a lithium iron phosphate battery pack, or a capacitor energy storage unit, and it is equipped with power monitoring, overcharge and over-discharge protection, and temperature control management functions. At the same time, the off-grid energy storage battery 12 can not only provide power to the actuators and sensor elements 13 of the photovoltaic platform 11, but also serve as a training device to simulate the off-grid operation scenario of an independent power station during training.

[0041] The sensor element 13 is installed in the photovoltaic power generation unit 11A and the photovoltaic power generation area according to the designer's plan. The sensor element 13 includes, but is not limited to, light intensity sensor 13A, temperature sensor 13B, humidity sensor 13C, wind speed sensor 13D and air quality sensor 13E.

[0042] Among them, the light sensor 13A is used to detect real-time solar irradiance and provide a reference for calculating the tracking angle; the temperature sensor 13B and humidity sensor 13C are used to detect changes in ambient temperature and humidity and provide data for battery management and experimental teaching; the wind speed sensor 13D is used to detect ambient wind speed data; and the air quality sensor 13E is used for air quality index monitoring.

[0043] The control device 14 connects to and controls the photovoltaic platform 11, the off-grid energy storage battery 12, and the sensor element 13 to execute the photovoltaic power generation dual-axis tracking and off-grid control method 20 in this application.

[0044] In some examples of this application, the control device 14 can also interact with a host computer or mobile terminal via wireless communication, which facilitates remote monitoring and data collection by operators, and supports parameter recording and automatic report generation during experimental teaching.

[0045] In some embodiments of this application, the system 10 further includes a photovoltaic cleaning element 15 disposed in the region of the photovoltaic power generation unit 11A for cleaning the surface of the photovoltaic power generation unit 11A by at least one of vibration, brushing, air blowing and electromagnetic tapping.

[0046] Specifically, if vibration is used for cleaning, an eccentric mass vibrator or piezoelectric vibrator is installed on the back of each photovoltaic power generation unit 11A or on the bracket. When the eccentric mass vibrator or piezoelectric vibrator is powered on, it generates high-frequency vibration, which causes the dust to fall off due to inertia.

[0047] If a brush cleaning method is used, a linear driver and a soft brush are installed on each photovoltaic power generation unit 11A. During cleaning, the linear driver drives the connected soft brush to extend and sweep away the dust on the surface of the photovoltaic power generation unit 11A.

[0048] If air blowing is used for cleaning, a blower array connected to a compressed air source is installed on each photovoltaic power generation unit 11A. The blower array faces the surface of the photovoltaic power generation unit 11A. During cleaning, compressed air connected to the compressed air source is blown onto the surface of the photovoltaic power generation unit 11A.

[0049] If electromagnetic impact is used, an electromagnetic impactor is installed on the support of each photovoltaic power generation unit 11A. The electromagnetic impactor uses short pulse vibration to remove dust from the photovoltaic power generation unit 11A.

[0050] In practice, two or more of the following methods can be used in combination: vibration, brushing, air blowing, and electromagnetic tapping, to improve cleaning efficiency.

[0051] Therefore, for reference Figure 6 As shown, Figure 6 A schematic flowchart illustrating an exemplary photovoltaic power generation dual-axis tracking and off-grid control method according to some embodiments of this application is shown. This application also relates to a photovoltaic power generation dual-axis tracking and off-grid control method 20, which includes:

[0052] 21: Acquire sunlight data, weather forecast data, satellite cloud image data, and environmental parameters to form multimodal photovoltaic data.

[0053] Specifically, in step 21, the illumination data is collected by illumination sensors deployed on the surface of the photovoltaic power generation unit and in the photovoltaic power generation area. These sensors can be silicon photovoltaic cell type or photodiode type sensors to obtain the current solar irradiance in real time. The weather forecast data is obtained through interaction with an external meteorological service platform via a wireless network. This weather forecast data includes at least the solar irradiance forecast, cloud cover distribution, wind speed, and temperature trends for the next few hours or days. The satellite cloud image data is obtained by accessing satellite image data from a meteorological satellite monitoring platform via a wireless network. After acquiring the raw satellite cloud image, it undergoes denoising, grayscale conversion, and edge detection processing. Then, existing image recognition algorithms are used to identify cloud distribution areas and movement trends, and the processed satellite cloud image features are stored in the form of numerical cloud occlusion parameters. The environmental parameters include temperature, humidity, wind speed, and air quality parameters, which are collected by temperature sensors, humidity sensors, wind speed sensors, and air quality sensors.

[0054] During the data collection process, the average, peak, and effective values ​​of all collected data were calculated to eliminate the interference of instantaneous fluctuations. At the same time, the timestamps of all collected data were aligned to ensure that the data were on a unified time base.

[0055] 22: Based on multimodal photovoltaic data, a light prediction model is constructed using machine learning methods to output the target solar irradiance and the target tracking angle;

[0056] Specifically, refer to Figure 7 As shown, Figure 7 The diagram illustrates an exemplary method for constructing an illumination prediction model according to some embodiments of this application. Step 22, constructing the illumination prediction model, includes:

[0057] 22A: Use linear regression or long short-term memory networks to model time series of weather forecast data and satellite cloud image data to predict future illumination trends over a predetermined time range.

[0058] Specifically, the weather forecast data includes future hourly or daily trends in solar irradiance, cloud cover distribution, wind speed, and temperature. After image processing and feature extraction, the satellite cloud image data is converted into numerical features such as cloud occlusion rate, cloud movement speed, and direction. Thus, the constructed illumination prediction model takes the aforementioned time series data as input and outputs the illumination trend curve within a specified future time window.

[0059] In practice, Long Short-Term Memory (LSTM) networks can capture temporal dependencies and are suitable for predicting illumination under complex weather conditions, while linear regression methods are suitable for scenarios with limited computing resources or those requiring rapid estimation; the specific settings are determined by the designers based on actual needs and costs.

[0060] 22B: Input the actual illumination data and environmental parameters into the correction layer to dynamically correct the illumination trend prediction results and generate the corrected illumination trend prediction value.

[0061] Specifically, the actual illumination data, temperature, humidity, wind speed, and air quality index data obtained by the illumination sensor are input into the correction layer of the illumination prediction model. Then, the correction layer uses a weighted average method, Kalman filtering, or a lightweight neural network based on residual learning to dynamically correct the prediction results of the illumination prediction model in real time, so as to compensate for the model deviation caused by factors such as local shading, dust pollution, and haze, thereby generating a corrected illumination trend prediction value.

[0062] Therefore, in some examples of this application, if a linear regression method is used to model the illumination prediction model, the prediction formula can be, for example, as follows: ,in For the future time Predicted solar irradiance intensity, For linear regression, Let be the regression coefficient of the i-th feature. The i-th input feature value at time t includes features such as cloud cover percentage, air humidity, air temperature, cloud obstruction rate extracted from satellite cloud images, and cloud movement speed and direction.

[0063] Furthermore, the correction layer of this illumination prediction model uses a weighted average correction formula, namely:

[0064] ,in This is the corrected predicted light intensity value. The actual light intensity measured in real time by the light sensor; α is a weighting coefficient with a value range of 0 ≤ α ≤ 1. α can be dynamically adjusted based on the air quality index (AQI / PM2.5) and wind speed.

[0065] Understandably, to eliminate the impact of differences in physical dimensions and numerical scales between different input features on the model, the input features are normalized or standardized before being introduced into the model, transforming them into dimensionless variables. Specifically, the input features include cloud cover percentage, air humidity, air temperature, wind speed, and parameters such as cloud cover rate, cloud movement speed, and direction extracted from satellite cloud images. Since these parameters have different physical dimensions and value ranges, directly participating in linear weighted calculations would lead to a lack of comparability in the model coefficients and affect model stability; therefore, a uniform scale transformation is required.

[0066] In some embodiments of this application, a minimum-maximum normalization method can be used to map each input feature to a preset interval (e.g., the [0,1] interval). The processing method is as follows: the minimum and maximum values ​​of each feature are obtained by statistical analysis based on historical data, and the current input feature is linearly scaled accordingly.

[0067] In some embodiments of this application, a standardization method may also be used to convert the input features into dimensionless variables with a mean of 0 and a standard deviation of 1.

[0068] After the above normalization or standardization process, each input feature is converted into a dimensionless value, thereby ensuring that the weighted summation operation in the linear regression model remains consistent in terms of dimensions and improving the convergence and prediction stability of the model training process.

[0069] Furthermore, after the model outputs the predicted solar irradiance, the predicted result can be restored to a solar irradiance value with actual physical meaning according to the corresponding inverse transformation rules used in the normalization or standardization process.

[0070] Therefore, in some examples of this application, if a long short-term memory network is used to model the illumination prediction model, then, for example, the calculation formula for the forget gate is: The formula for calculating the input gate is: The formula for calculating candidate memories is: The formula for calculating memory state updates is: The formula for calculating the output gate is: The formula for calculating the hidden state is: ;in, Let be the input vector at time t. These are the activation values ​​for the forget gate, input gate, and output gate, respectively. For the memory state at time t, Let be the hidden state at time t. For the hidden state at time t-1, This is the weight matrix. Here, sigmoid is the activation function, and tanh is the hyperbolic tangent activation function. For the forget gate bias term, For the input gate bias term, For candidate memory bias terms, This is the output gate bias term.

[0071] It is understandable that the input data of the illumination prediction model constructed using a long short-term memory network is usually normalized or standardized before being used for model training and inference calculations. Therefore, the model output is a dimensionless value corresponding to the normalized space. In order to make the prediction results have actual physical meaning, after obtaining the output of the neural network, the output results are subjected to inverse normalization processing to map them from the dimensionless space back to the original physical quantity space, thereby obtaining the solar irradiance intensity value with actual physical dimensions.

[0072] Therefore, the final formula for calculating the predicted light intensity is: ,in For the future time Predicted solar irradiance intensity, For output layer weights, This is the output layer bias term.

[0073] Furthermore, residual correction is used in the correction layer of this illumination prediction model: ,in, For residual correction at time t; This represents the dynamic correction value based on the real-time error. For lightweight neural networks or Kalman filters, The actual solar irradiance measured at time t;

[0074] Understandably, residual correction is used to dynamically compensate the output of the illumination prediction model. It is expressed as adding a dynamic correction value to the initial predicted solar irradiance. To ensure the consistency of this correction process in a physical sense, the dynamic correction value and the predicted solar irradiance have the same physical dimensions, which are both radiation power per unit area. Therefore, when performing the addition operation, the dimensions of each item remain consistent, thus ensuring that the corrected result is still a solar irradiance with actual physical meaning.

[0075] 22C: Calculate the target solar irradiance and target tracking angle by combining the geographical location and time information of the photovoltaic platform.

[0076] Specifically, the theoretical solar position parameters (solar altitude angle and azimuth angle) are calculated using geographical location and time information. Then, the corrected illumination trend prediction value is combined with the theoretical solar position parameters to obtain the target solar irradiance. Finally, based on the directional distribution of the target solar irradiance, the target tracking angle (azimuth angle and elevation angle) is calculated.

[0077] 23: Control the photovoltaic platform to rotate as a whole according to the target tracking angle and make local independent adjustments to multiple independently adjustable photovoltaic power generation units to achieve a composite tracking action sequence.

[0078] Specifically, refer to Figure 8 As shown, Figure 8 The diagram illustrates an exemplary flowchart of a method for implementing a composite light-tracking action sequence according to some embodiments of this application. Step 23, implementing the composite light-tracking action sequence, includes:

[0079] 23A: Compare the target tracking angle with the current attitude of the photovoltaic platform and determine whether there is a deviation based on the comparison result. If so, proceed to step 23B: Calculate the overall adjustment compensation amount of the photovoltaic platform and control the overall rotation of the photovoltaic platform based on the overall adjustment compensation amount.

[0080] Specifically, the output target tracking angle is compared with the current attitude angle of the photovoltaic power generation unit of the photovoltaic platform. If the difference between the two is less than the set threshold, the overall deviation is considered negligible, the overall rotation is skipped, and only local fine-tuning is performed; if the deviation exceeds the set threshold, then step 23B is performed.

[0081] In step 23B, the overall adjustment compensation amount is calculated based on the difference between the target tracking angle and the current attitude angle, and corresponding control commands are generated to drive the overall dual-axis rotation mechanism of the platform, so that the entire photovoltaic platform can be quickly adjusted to near the target angle. During the overall rotation, acceleration-deceleration curve control can be used to reduce motor impact. If the ambient wind speed exceeds the set upper limit, the rotation speed can be automatically reduced or the overall rotation can be paused to ensure structural safety.

[0082] 23C: Based on the local illumination data and target tracking angle of each photovoltaic power generation unit on the photovoltaic platform, calculate the independent compensation amount for each photovoltaic power generation unit and control the corresponding photovoltaic power generation unit to adjust independently according to the independent compensation amount.

[0083] Specifically, after step 23B is completed, that is, after the overall rotation is completed, the local illumination data of each photovoltaic power generation unit of the photovoltaic platform is compared with the target tracking angle, and the independent compensation amount of each photovoltaic power generation unit is calculated. Proportional control or fuzzy control function can be used in the calculation. Then, based on the calculation results, each photovoltaic power generation unit is driven to make independent fine adjustments so that its attitude angle is closer to the target angle, thereby realizing compensation under local illumination differences.

[0084] For example, when the overall deviation is greater than 5°, overall rotation and local adjustment are performed; when the overall deviation is less than or equal to 5°, only local adjustment is performed.

[0085] For example, when light intensity changes rapidly (such as when clouds pass by), overall rotation is prioritized to quickly follow the sun; when light intensity is stable, local fine-tuning is prioritized to save energy. In cases where wind speed exceeds a set threshold or during sandstorms, overall rotation is temporarily locked, with only small local adjustments maintained to reduce mechanical wear.

[0086] 24: Predictive data is shared among multiple photovoltaic platforms via wireless communication. When the lead photovoltaic platform detects that the change in illumination exceeds a preset threshold, a correction command is generated and sent to other photovoltaic platforms, enabling them to correct the tracking sequence in advance based on the received correction command and the frequency of illumination change.

[0087] Specifically, refer to Figure 9 As shown, Figure 9 A flowchart illustrating an exemplary multi-photovoltaic platform collaborative method according to some embodiments of this application is shown. Step 24, which involves pre-correcting the light-tracking action sequence, includes:

[0088] 24A: Multiple photovoltaic platforms share illumination data, satellite cloud imagery data, and environmental parameters via wireless communication.

[0089] Specifically, multiple photovoltaic platforms share data through wireless communication networks (such as LoRa, ZigBee, Wi-Fi Mesh, or 5G private networks). The shared data includes at least real-time irradiance data, satellite cloud image data fragments (such as compressed cloud coverage image features), environmental parameters (temperature, humidity, air quality index, wind speed, etc.), and local prediction results (target irradiance and target tracking angle). The data sharing method can be broadcast (all platforms communicate with each other) or distributed hierarchical (the pilot platform distributes the data, and other platforms receive it).

[0090] 24B: Select a photovoltaic platform located in a leading position for illumination as the leading photovoltaic platform.

[0091] Specifically, in group collaboration, photovoltaic platforms located in the light-leading position are preferentially selected as lead photovoltaic platforms. The light-leading position refers to the platform that first comes into contact with changes in light intensity (such as cloud cover or increased light intensity) on the sun's path. The lead photovoltaic platform can be determined by comparing the rate of change of light intensity of each platform. That is, if a photovoltaic platform detects that the rate of decrease or increase in light intensity exceeds a set threshold, then that photovoltaic platform is selected as the lead photovoltaic platform.

[0092] In practice, the threshold for light intensity can be dynamically adjusted based on the season, geographical location, or weather pattern.

[0093] 24C: When the pilot photovoltaic platform detects that the change in light intensity exceeds the preset threshold, it predicts the target tracking angle based on the current location's light intensity data, satellite cloud image data, and environmental parameters, and generates a correction command based on the target tracking angle and sends it to other photovoltaic platforms.

[0094] Specifically, when the pilot photovoltaic platform detects that the change in light intensity exceeds a preset threshold, it combines real-time light information of the current location, satellite cloud image, and surrounding environmental parameters to predict the light trend and target tracking angle for the next time period. Then, based on the difference between the predicted angle and the current angle, it generates a correction command and transmits it to other photovoltaic platforms in the form of a communication data packet.

[0095] In practical implementation, a weighted linear prediction model can be used for prediction, thus setting the time as... The prediction window length is The predicted next moment The light intensity is :

[0096] ,in To measure the real-time light intensity at the current location of the pilot photovoltaic platform, The reference irradiance at the target time is calculated from satellite cloud images. These are the surrounding environmental parameters (such as cloud cover, temperature, humidity, wind speed, etc.). This is a mapping function from environmental parameters to the influence of illumination. For the weighting coefficients, satisfying It uses recursive least squares or adaptive Kalman gain methods to adaptively adjust based on historical errors.

[0097] Understandably, to ensure the consistency of the mapping process in a physical sense, the output of the mapping function is limited to solar irradiance or its corresponding normalized value, so that it can be fused with other predicted components; all features are normalized or standardized before entering the model, so that they are transformed into dimensionless variables.

[0098] In actual implementation, the target tracking angle is expressed as azimuth (yaw). With pitch angle Based on the principle of maximizing illumination, the target angle is predicted. This can be obtained by maximizing the sun's position or the illumination gradient. A first-order linear approximation can be used:

[0099] , ,in The predicted azimuth angle, For the predicted pitch angle, This is the actual azimuth angle. This is the actual pitch angle. Based on the above The calculated angle suggests a correction amount. This is the angle mapping coefficient used to prevent overshoot.

[0100] This generates the error amount for the correction instruction: , The correction instructions are then packaged into communication data packets and transmitted to other photovoltaic platforms in the form of communication data packets; among them, This is the correction amount for the azimuth angle. It is a correction for the pitch angle, measured in degrees (e.g., degrees or radians).

[0101] If other platforms receive correction instructions and observe that the frequency of changes in illumination exceeds the advance adjustment frequency threshold, then the action sequence will be executed in advance according to the instructions, expressed as a time advance: ,in To adjust the factor, To predict confidence levels, This is the maximum allowable lead time. Therefore, from a practical application perspective, it is modified as follows: , ,in For the execution ratio, For azimuth correction in practical applications, For pitch angle correction in practical applications.

[0102] Understandably, the adjustment factor is a dimensionless parameter used to adjust the advance level according to the system's operating status or empirical strategies. Its value range can be set between 0 and 1 or configured according to actual needs. The prediction confidence is used to characterize the reliability of the current illumination prediction result, and is usually a dimensionless probability value between 0 and 1. The maximum advance duration is a preset time parameter, and its dimension is time (e.g., seconds or minutes), used to limit the upper limit of the advance amount. Thus, the dimension of the time advance amount is determined by the maximum advance duration, and the overall calculation result is still a time quantity, thereby ensuring the consistency of the formula in terms of dimension.

[0103] All features are normalized or standardized before being entered into the model, transforming them into dimensionless variables.

[0104] 24D: Other photovoltaic platforms can adjust the tracking sequence in advance according to the received correction instructions and the frequency of changes in light intensity, without waiting for local light conditions to fluctuate.

[0105] In actual implementation, when encountering rapidly moving clouds or frequent fluctuations in sunlight, the photovoltaic platform can prioritize overall rotation compensation and complete the synchronous correction of azimuth and pitch angles in a short period of time.

[0106] If the changes in illumination are relatively gradual, the overall number of rotations can be reduced by making only local adjustments, thereby reducing drive energy consumption.

[0107] Therefore, while performing advance corrections, the photovoltaic platform can also perform secondary corrections by combining local real-time measurement data to avoid over-adjustment due to deviations in remote forecasts.

[0108] In some examples of this application, swarm intelligence optimization methods, such as ant colony optimization or particle swarm optimization, can be introduced when correcting the action sequence in advance to dynamically coordinate the light-tracking actions of multiple photovoltaic platforms.

[0109] If the ant colony algorithm is used, each photovoltaic platform can be regarded as an individual ant, and its angle adjustment path corresponds to the ant's path selection. The photovoltaic platforms form pheromones by sharing correction instructions, and the overall coordination is gradually improved after iterative updates.

[0110] If the particle swarm optimization algorithm is used, each photovoltaic platform is abstracted as a particle, and the fitness function is constructed using the difference between the attitude angle and the predicted angle. The algorithm iterates between the global and local optima, so that multiple photovoltaic platforms gradually approach a better light-tracking sequence.

[0111] The above technical solutions can not only improve the accuracy of illumination prediction, but also reduce energy waste caused by ineffective adjustments.

[0112] In a collaborative control mode, once the lead photovoltaic platform detects a rapid fluctuation in illumination, it can generate a correction command first, allowing the other photovoltaic platforms to adjust their actions before they even perceive a significant change in illumination. This shortens response time and avoids the delays and errors that might result from independent control of a single photovoltaic platform.

[0113] refer to Figure 10 As shown, Figure 10 The diagram illustrates a flowchart of an exemplary method for dynamically adjusting a light-tracking strategy according to some embodiments of this application. In some embodiments of this application, the method further includes: 201: Real-time detection of the state of charge of the off-grid energy storage battery.

[0114] Specifically, by collecting battery voltage, current and temperature data, the state of charge (SOC) is calculated using the Coulomb Counting method and the Open-Circuit Voltage (OCV) method. The calculated SOC results are then compensated for by incorporating battery temperature to avoid the influence of temperature on capacity estimation.

[0115] 202: Dynamically adjust the solar tracking strategy based on the energy storage state of charge. The dynamic adjustment of the solar tracking strategy includes:

[0116] When the energy storage is close to full charge, it enters a low-activity energy-saving mode;

[0117] When energy storage is insufficient, enhance solar tracking to maximize power generation.

[0118] In the low-motion energy-saving mode, the adjustment frequency is reduced so that the overall rotation is only performed when the deviation is greater than 5°, and the local independent fine adjustment is only triggered when the light intensity drops beyond the set threshold, thereby reducing motor operation.

[0119] In enhancing beam tracking to maximize power generation, the frequency of overall rotation and local adjustment is increased. For example, the deviation threshold is reduced to 1°, and the beam tracking action is controlled to prioritize the angle with the best power generation efficiency, even if the motor energy consumption increases.

[0120] Specifically, in step 202, when dynamically adjusting the light-tracking strategy based on the energy storage state of charge, the method further includes:

[0121] 202A: Determine the state of charge of each off-grid energy storage battery.

[0122] In 202A, the system periodically reads the state of charge of the off-grid energy storage battery and compares the read value with a set threshold range. A first threshold and a second threshold can be set. For example, the first threshold can be set between 85% and 95%, and the second threshold can be set between 20% and 30%.

[0123] 202B: If the state of charge of the off-grid energy storage battery is greater than the first threshold, control the corresponding photovoltaic platform of the off-grid energy storage battery to enter the low-action energy-saving mode to reduce the number of angle adjustments.

[0124] When the state of charge of the off-grid energy storage battery is greater than the first threshold, it means that the off-grid energy storage battery is close to full charge. At this time, the adjustment frequency is reduced and the overall rotation is only executed when the deviation is greater than 5°. Local independent fine adjustment is only triggered when the light intensity drops beyond the set threshold, thereby reducing the motor operation.

[0125] 202C: If the state of charge of the off-grid energy storage battery is less than the second threshold, control the corresponding photovoltaic platform of the off-grid energy storage battery to enter the high-efficiency power generation mode and increase the angle adjustment frequency.

[0126] When the state of charge of the off-grid energy storage battery is less than the first threshold, it means that the off-grid energy storage battery is not charged enough. At this time, the frequency of overall rotation and local adjustment is increased, for example, the deviation threshold is reduced to 1°, and the tracking action is controlled to prioritize the angle with the best power generation efficiency, even if the motor energy consumption increases.

[0127] 202D: If the state of charge of the off-grid energy storage battery is between the first threshold and the second threshold, the corresponding photovoltaic platform of the off-grid energy storage battery is controlled to maintain the conventional tracking mode, so as to balance and adjust the composite tracking action sequence according to the target tracking angle output by the prediction model.

[0128] When the state of charge of the off-grid energy storage battery is between the first threshold and the second threshold, it means that the off-grid energy storage battery is in the normal operating range. At this time, the photovoltaic platform executes a composite tracking action sequence according to the target tracking angle output by the prediction model. The overall rotation and local adjustment actions are carried out at a normal rhythm to achieve a balance between power generation and energy consumption.

[0129] In actual implementation, the above-mentioned light-tracking strategy can be adjusted independently for each off-grid energy storage battery and photovoltaic power generation unit; among them, the first threshold and the second threshold can be dynamically adjusted according to the season, load demand and battery degradation status.

[0130] refer to Figure 11 As shown, Figure 11 A flowchart illustrating an exemplary exception handling method according to some embodiments of this application is shown. In some embodiments of this application, when the photovoltaic platform executes a sequence of light-tracking actions, the method further includes:

[0131] 203: By comparing the power output of the photovoltaic power generation unit with the expected value of the prediction model, it is determined whether there is a fault in the photovoltaic sub-unit; at the same time, the data collected by the sensor elements is calibrated to determine whether the data drifts or exceeds the safety threshold.

[0132] Specifically, during the process of tracking sunlight on the photovoltaic platform, the real-time output power of each photovoltaic power generation unit is acquired in real time and compared with the expected power calculated by the irradiance prediction model based on the target irradiance. If the comparison result continuously exceeds or falls below the set threshold, it is determined that the photovoltaic power generation unit has abnormal performance or is partially shaded.

[0133] Meanwhile, the measured values ​​of each sensor element collected in real time are compared with the reference values. If the comparison result continues to exceed the drift threshold, it is determined that the sensor element is drifting or malfunctioning. The reference values ​​can come from similar sensor elements in adjacent photovoltaic power generation units, or data shared collaboratively by the group.

[0134] In actual implementation, a safety threshold is also set. If the collected sensor data exceeds the set physical safety range (e.g., temperature > 85℃), a protection mechanism is triggered to prevent incorrect light tracking. For example, this protection mechanism can replace the data exceeding the safety threshold with data from similar sensor elements of other photovoltaic power generation units adjacent to the photovoltaic power generation unit.

[0135] If so, proceed to step 204: identify the photovoltaic platform corresponding to the fault or sensor element drift as an abnormal photovoltaic platform, and automatically correct the tracking angle of the abnormal photovoltaic platform, or switch the abnormal photovoltaic platform to redundant operation mode.

[0136] Specifically, when an anomaly is detected in a photovoltaic (PV) power generation unit or sensor element, if the output power of a certain PV power generation unit decreases but has not completely failed, the tracking angle of the PV power generation unit can be fine-tuned to observe whether the power output shows a recovery trend. If it is determined that the PV power generation unit or associated sensor element has completely failed, the system can switch to a redundant operation mode. In this redundant operation mode, the PV power generation unit no longer performs active tracking, but only maintains a pre-set safety angle, such as a horizontal or slightly tilted protective angle, to reduce the load on the structure and avoid secondary damage. At the same time, the output data of the PV power generation unit is automatically removed and no longer included in the global control calculation. Instead, the measurement results of adjacent normal PV power generation units or adjacent sensor elements are called for compensation.

[0137] In some examples of this application, when any photovoltaic power generation unit or sensor element enters an abnormal state, a group coordination strategy can be activated to obtain shared predicted values ​​or historical data from neighboring photovoltaic power generation units to replace the data input of the abnormal photovoltaic power generation unit, thereby maintaining the stable execution of the overall light-tracking strategy.

[0138] In practical implementation, anomalies can be categorized into minor anomalies, severe anomalies, and emergency anomalies. For example, minor anomalies only require minor angle adjustments, severe anomalies necessitate switching to redundancy mode, and emergency anomalies require immediate shutdown for protection. Furthermore, the judgment threshold can be adjusted based on the power plant scale and operation and maintenance strategy. Upon entering redundancy mode, the fault status can be checked once or multiple times according to a preset detection cycle, such as 5 minutes to 2 hours. Once the fault is identified and the system returns to normal, it automatically exits redundancy mode and reconnects to the tracking beam control.

[0139] Therefore, through the above-mentioned anomaly detection and redundant operation technology, even if some photovoltaic power generation units or sensor components fail, the photovoltaic platform as a whole can still maintain relatively stable operation, reducing the risk of large-scale failure caused by a single point of failure.

[0140] refer to Figure 12 As shown, Figure 12 A schematic flowchart of an exemplary periodic photovoltaic cleaning method according to some embodiments of this application is shown. In some embodiments of this application, when the photovoltaic platform performs a sequence of light-tracking actions, the method further includes:

[0141] 205: Every first preset time interval, the photovoltaic platform is adjusted from the normal power generation posture to the preset clean posture, wherein after the photovoltaic platform is adjusted to the preset clean posture, the photovoltaic panels of the photovoltaic platform are at least partially facing the clean area.

[0142] Specifically, the first preset time can be set according to the operator's settings or the local dust accumulation, such as between 20:00 and 4:00 every night, or every 12 hours to 7 days; the preset cleaning posture can be set to a specific tilt angle so as to use gravity to help the dust slide down and provide a favorable angle for cleaning actions such as contact or water spraying of photovoltaic cleaning elements.

[0143] In actual implementation, if wind speed is detected Or precipitation / snowfall indicator When necessary, the cleaning task can be automatically postponed or paused. For instantaneous wind speed, This is a safety threshold (e.g., 15 m / s). The precipitation / snowfall indicator (derived from sensor or meteorological input, 0 / 1) ensures the structural safety of the tracking platform under abnormal environments and prevents damage caused by severe weather.

[0144] 206: After the photovoltaic platform is adjusted to the preset cleaning posture, the corresponding cleaning unit enters the working state of the second preset time (e.g., set to any time in the range of 3 to 30 minutes) (e.g., including one or more of vibration, brushing, air blowing or electromagnetic tapping), removes the stains on the surface of the photovoltaic module during this period, and restores the photovoltaic platform posture to the normal power generation state after cleaning is completed.

[0145] In actual implementation, the second preset time is set as follows: Use inventory model to dynamically adjust For example: ,in In time The second preset time is adopted. Baseline cleaning time, This is an indicator of dust accumulation rate (which can be determined by commissioning personnel based on optical transmittance or the rate of decrease in shortwave power generation efficiency). Air Quality Index. This represents the upper dimension of AQI. This is the adjustment coefficient; The cutting factor is used to make exist scope; The minimum value of the second preset time. This is the second preset maximum time value.

[0146] Furthermore, seasonal factors can be introduced for seasonal parameters (such as dry and dusty seasons, rainy and snowy seasons). It can be applied to either the second preset time or the cleaning frequency. : ,in This is the baseline cleaning frequency. To adjust the coefficient (e.g., dry season) (Then increase the frequency).

[0147] Understandably, all features in this application are normalized or standardized before calculation, transforming them into dimensionless variables.

[0148] In some examples of this application, the first preset time and the second preset time can be dynamically adjusted according to the decrease in light intensity, the rate of dust accumulation, or the air quality index.

[0149] For example, the cleaning cycle can be shortened or the cleaning time increased during dry and dusty seasons; the cleaning cycle can be extended or the cleaning time reduced during rainy and snowy seasons. Alternatively, the cleaning cycle can be extended and the cleaning time increased during rainy and snowy seasons, and the specific settings can be made by the commissioning personnel according to actual needs.

[0150] Therefore, by setting up a periodic self-cleaning system for the photovoltaic platform, the surface of the photovoltaic power generation unit can be kept clean, improving the light incident efficiency and reducing power generation loss caused by dust or pollutants. This is suitable for unattended or off-grid photovoltaic systems.

[0151] Therefore, the input features of the machine learning model involved in this application are normalized or standardized before entering the model to eliminate the influence of different physical dimensions; the model output is converted into solar irradiance with actual physical meaning through inverse normalization; the output of each correction term, residual term and mapping function has the same dimension as the target physical quantity.

[0152] In some embodiments, reference Figure 13 As shown, Figure 13 A connection diagram of an electronic device used to implement embodiments of this application is shown. The electronic device 30 includes a memory 31 and a processor 32. The memory 31 stores a computer program that can run on the processor 32. When the processor 32 executes the computer program, it implements the methods described in the above embodiments. The number of memories 31 and processors 32 can be one or more.

[0153] The electronic device 30 also includes:

[0154] Communication interface 33 is used to communicate with external devices and perform data exchange and transmission.

[0155] If the memory 31, processor 32, and communication interface 33 are implemented independently, then the memory 31, processor 32, and communication interface 33 can be interconnected through a bus and complete communication between them.

[0156] This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 13 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0157] Optionally, in a specific implementation, if the memory 31, processor 32, and communication interface 33 are integrated on a single chip, then the memory 31, processor 32, and communication interface 33 can communicate with each other through an internal interface.

[0158] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor 32, implements the method provided in this application.

[0159] This application also provides a chip, which includes a processor 32 for calling and running instructions stored in a memory 31, causing a communication device equipped with the chip to execute the method provided in this application.

[0160] This application also provides a chip, including: an input interface, an output interface, a processor 32, and a memory 31. The input interface, the output interface, the processor 32, and the memory 31 are connected through an internal connection path. The processor 32 is used to execute code in the memory 31. When the code is executed, the processor 32 is used to execute the method provided in the application embodiment.

[0161] It should be understood that the processor 32 mentioned above can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that processor 32 can be a processor supporting the Advanced Reduced Instruction Set Computing (ARM) architecture.

[0162] Furthermore, the aforementioned memory 31 may include read-only memory and random access memory, and may also include non-volatile random access memory. The memory 31 may be volatile memory or non-volatile memory, or may include both. The non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may include random access memory (RAM), which serves as an external cache. Many forms of RAM are available by way of example, but not limitation. Examples include Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).

[0163] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0164] Other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and this application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for dual-axis tracking and off-grid control of photovoltaic power generation, characterized in that, The method includes a photovoltaic platform equipped with multiple independently adjustable photovoltaic power generation units, comprising: Acquire solar illumination data, weather forecast data, satellite cloud image data, and environmental parameters to generate multimodal photovoltaic data; Based on the multimodal photovoltaic data, a light prediction model is constructed using machine learning methods, which outputs the target solar irradiance and the target tracking angle. The photovoltaic platform is rotated as a whole and multiple independently adjustable photovoltaic power generation units are locally adjusted according to the target tracking angle, thereby realizing a composite tracking action sequence; and Predictive data is shared among multiple photovoltaic platforms via wireless communication. When the lead photovoltaic platform detects that the change in illumination exceeds a preset threshold, a correction command is generated and sent to other photovoltaic platforms, enabling them to adjust their tracking sequence in advance based on the received correction command and the frequency of illumination change.

2. The method according to claim 1, characterized in that, The construction of the illumination prediction model includes: The time series of weather forecast data and satellite cloud image data are modeled using linear regression or long short-term memory networks to predict the future illumination trend within a preset time range; The actual illumination data and environmental parameters are input into the correction layer to dynamically correct the illumination trend prediction results and generate the corrected illumination trend prediction value. By combining the geographical location and time information of the photovoltaic platform, the target solar irradiance intensity and the target tracking angle are calculated.

3. The method according to claim 1 or 2, characterized in that, The sequence of actions to achieve composite light tracking includes: The target tracking angle is compared with the current attitude of the photovoltaic platform, and the deviation is determined based on the comparison result. If there is a deviation, the overall adjustment compensation amount of the photovoltaic platform is calculated, and the overall rotation of the photovoltaic platform is controlled according to the overall adjustment compensation amount. Based on the local illumination data and target tracking angle of each photovoltaic power generation unit on the photovoltaic platform, the independent compensation amount of each photovoltaic power generation unit is calculated, and the corresponding photovoltaic power generation unit is independently adjusted according to the independent compensation amount.

4. The method according to claim 3, characterized in that, The advance correction tracking action sequence includes: Multiple photovoltaic platforms share solar illumination data, satellite cloud imagery data, and environmental parameters via wireless communication. A photovoltaic platform located at the leading position of sunlight was selected as the pilot photovoltaic platform; When the pilot photovoltaic platform detects that the change in light intensity exceeds a preset threshold, it predicts the target tracking angle based on the current location's light intensity data, satellite cloud image data, and environmental parameters, and generates a correction command based on the target tracking angle and sends it to other photovoltaic platforms. Other photovoltaic platforms adjust their tracking sequence in advance based on the received correction instructions and the frequency of changes in sunlight.

5. The method according to claim 4, characterized in that, When correcting the tracking sequence in advance, the tracking sequence of multiple photovoltaic platforms is dynamically optimized using ant colony optimization or particle swarm optimization.

6. The method according to claim 1 or 5, characterized in that, It also includes off-grid energy storage batteries, and the method further includes: Real-time monitoring of the state of charge of off-grid energy storage batteries; The dynamic adjustment of the light-tracking strategy based on the energy storage state of charge includes: When the energy storage is close to full charge, it enters a low-activity energy-saving mode; When energy storage is insufficient, enhance solar tracking to maximize power generation.

7. The method according to claim 6, characterized in that, When monitoring the state of charge of off-grid energy storage batteries in real time, the method further includes: Determine the state of charge of each off-grid energy storage battery; If the state of charge of the off-grid energy storage battery is greater than the first threshold, the corresponding photovoltaic platform of the off-grid energy storage battery is controlled to enter a low-action energy-saving mode to reduce the number of angle adjustments. If the state of charge of the off-grid energy storage battery is less than the second threshold, the corresponding photovoltaic platform of the off-grid energy storage battery is controlled to enter the high-efficiency power generation mode and the angle adjustment frequency is increased. If the state of charge of the off-grid energy storage battery is between the first threshold and the second threshold, the corresponding photovoltaic platform of the off-grid energy storage battery is controlled to maintain the conventional tracking mode, so as to balance and adjust the composite tracking action sequence according to the target tracking angle output by the prediction model.

8. The method according to claim 1 or 5, characterized in that, When the photovoltaic platform executes the light-tracking action sequence, the method further includes: By comparing the power output of the photovoltaic power generation unit with the expected value of the prediction model, it is determined whether there is a fault in the photovoltaic sub-unit; at the same time, the data collected by the sensor elements is calibrated to determine whether the data drifts or exceeds the safety threshold. If so, the photovoltaic platform corresponding to the fault or sensor component drift will be identified as an abnormal photovoltaic platform, and the tracking angle of the abnormal photovoltaic platform will be automatically corrected, or the abnormal photovoltaic platform will be switched to redundant operation mode.

9. The method according to claim 1 or 5, characterized in that, Including photovoltaic cleaning elements, the method further includes, when executing a light-tracking action sequence on a photovoltaic platform, the following: Every first preset time interval, the photovoltaic platform is adjusted from its normal power generation posture to a preset cleaning posture, with at least a portion of the surface of the photovoltaic platform facing the clean area. The photovoltaic cleaning element is controlled to enter a cleaning state for a second preset time to clean the contaminants on the surface of the photovoltaic platform. After cleaning is completed, the photovoltaic platform is adjusted from the preset cleaning posture back to the normal power generation posture.

10. A photovoltaic power generation dual-axis tracking and off-grid control system, characterized in that, include: The photovoltaic platform includes multiple independently adjustable photovoltaic power generation units and a first angle adjustment element, and each photovoltaic power generation unit is provided with at least one second angle adjustment element; An off-grid energy storage battery, which is connected to at least one photovoltaic power generation unit, for storing the energy collected by the connected photovoltaic power generation unit; Sensor elements, respectively installed in the photovoltaic power generation unit and the photovoltaic power generation area, are used to acquire light data and environmental parameters including light intensity, temperature, humidity, wind speed and air quality index of the target area; A photovoltaic cleaning element is disposed in a photovoltaic power generation unit and is used to clean the surface of the photovoltaic power generation unit by at least one method, such as vibration, brushing, air blowing and electromagnetic tapping. A control device, connected to and controlling the photovoltaic platform, off-grid energy storage battery, sensor elements, and photovoltaic cleaning elements, to perform the method according to any one of claims 1 to 9.

Images