Photovoltaic tracking bracket angle optimization and power generation efficiency improvement method based on weather forecast

Abstract

This invention provides a method for optimizing the angle of photovoltaic trackers and improving power generation efficiency based on weather forecasts. This method involves using meteorological data to construct a deep neural network model to predict future weather data, calculate the solar azimuth and altitude, establish a motion constraint model for the photovoltaic trackers to ensure safety and avoid risks, and use an improved dynamic programming algorithm to optimize and solve the optimal angle sequence, generating angle adjustment commands to control tracker adjustment. This method improves the power generation efficiency of photovoltaic modules while ensuring the safety of the system in inclement weather.

Description

Technical Field

[0001] The present invention relates to weather forecasting technology, and in particular to a method for optimizing the angle of a photovoltaic tracking bracket and improving power generation efficiency based on weather forecasting. Background Art

[0002] With the global energy restructuring and the rapid development of renewable energy, photovoltaic power generation, as a key component of clean energy, is gaining widespread application worldwide. To improve the efficiency of photovoltaic power plants, photovoltaic tracking technology has emerged. This technology adjusts the angle of photovoltaic modules to better track the sun's trajectory, maximizing the amount of solar radiation received, thereby increasing power generation.

[0003] Traditional photovoltaic tracking systems primarily employ simple tracking strategies based on time or light intensity, adjusting the tracker angle based on pre-calculated sun trajectory data or real-time light sensor information. These systems typically trigger angle adjustments based on fixed time intervals or changes in the sun's position, failing to fully consider the impact of weather changes on power generation efficiency. With the advancement of weather forecasting and artificial intelligence technologies, optimizing the tracker angle based on meteorological data has become a new approach to improving the efficiency of photovoltaic power plants.

[0004] Existing technologies generally lack the ability to accurately predict and respond to complex meteorological conditions. Most systems rely solely on historical data or simple weather forecasts, unable to make timely and accurate predictions and adjustments for extreme weather conditions such as short-term cloud changes and sudden changes in wind speed. This results in ineffective tracking strategies in unstable weather conditions.

[0005] Existing tracking systems lack effective safety mechanisms for extreme weather conditions. When encountering severe weather conditions such as strong winds and heavy rain, if the photovoltaic tracking brackets cannot be adjusted to a safe angle in a timely manner, equipment damage may occur, increasing maintenance costs and reducing the overall reliability and lifespan of the power station.

[0006] Traditional optimization algorithms are computationally inefficient and struggle to achieve a global optimum when solving for the optimal angle sequence for photovoltaic trackers. Most systems employ simple, immediate optimization strategies that fail to consider the overall goal of maximizing power generation throughout the day. As a result, while angle adjustments during certain time periods achieve immediate power maximization, they are suboptimal from a daily perspective, hindering the full potential of the photovoltaic power station. Summary of the Invention

[0007] The embodiments of the present invention provide a method for optimizing the angle of photovoltaic tracking brackets and improving power generation efficiency based on weather forecasting, which can solve the problems in the prior art.

[0008] A first aspect of an embodiment of the present invention provides a method for optimizing the angle of a photovoltaic tracking bracket and improving power generation efficiency based on weather forecasting, comprising:

[0009] Receiving meteorological data collected by a meteorological monitoring station for a target photovoltaic power station area, constructing a deep neural network model based on the meteorological data, and using the deep neural network model to predict meteorological data within a preset time period in the future to obtain predicted meteorological data;

[0010] Calculating the solar azimuth and solar altitude of the target photovoltaic power station area within the future preset time period based on the predicted meteorological data;

[0011] Based on the cloud coverage and wind speed data in the predicted meteorological data, a motion constraint model of the photovoltaic tracking bracket is established to determine whether the cloud coverage is greater than a preset constraint threshold and whether the wind speed data is greater than a safe wind speed threshold. When the judgment conditions are met, a safe risk avoidance angle of the photovoltaic tracking bracket is calculated, and the photovoltaic tracking bracket is controlled to rotate to the safe risk avoidance angle.

[0012] Based on the solar radiation intensity and temperature data in the predicted meteorological data, combined with the solar azimuth and the solar altitude angle, an improved dynamic programming algorithm is used to optimize and solve the optimal angle sequence of the photovoltaic tracking bracket within the future preset time period, wherein the optimal angle sequence is used to maximize the power generation efficiency of the photovoltaic assembly;

[0013] According to the optimal angle sequence, an angle adjustment instruction of the photovoltaic tracking bracket is generated, and the angle adjustment instruction is sent to a controller of the photovoltaic tracking bracket to control the photovoltaic tracking bracket to adjust the angle according to the optimal angle sequence.

[0014] A deep neural network model is constructed based on the meteorological data, and the deep neural network model is used to predict meteorological data within a preset time period in the future. The predicted meteorological data obtained includes:

[0015] Normalizing the meteorological data according to the adaptive time window length to obtain standardized features;

[0016] The deep neural network model includes a parallel feature extraction layer, an adaptive fusion layer, and a dynamic attention layer; the standardized features are input into the parallel feature extraction layer, and multi-scale local features are extracted through spatial pyramid convolution branches; the multi-scale local features are input into the adaptive fusion layer, adaptive weights are calculated based on feature discriminability, and feature fusion is performed to obtain enhanced features; the enhanced features are input into the dynamic attention layer, and key information at different time scales is captured based on a multi-scale temporal attention mechanism to obtain attention-enhanced features;

[0017] Constructing an adaptive multi-task loss function, the adaptive multi-task loss function including a prediction error loss term and a time series correlation loss term; adaptively adjusting the weight coefficients of each loss term based on the difficulty of the prediction task and the data distribution characteristics; and performing end-to-end training on the deep neural network model using the adaptive multi-task loss function;

[0018] The meteorological data to be predicted is input into the trained deep neural network model to generate predicted meteorological data.

[0019] Calculating the solar azimuth and solar altitude angle of the target photovoltaic power station area within the future preset time period based on the predicted meteorological data includes:

[0020] Performing time series decomposition on the predicted meteorological data to extract time-varying features and periodic features; performing data correction on the predicted meteorological data based on the time-varying features and the periodic features to obtain corrected predicted data;

[0021] Calculate the atmospheric refraction correction value based on the predicted pressure data, predicted temperature data and the site altitude in the corrected predicted data; calculate the atmospheric scattering correction value based on the predicted relative humidity data, predicted cloud cover data and the site altitude in the corrected predicted data;

[0022] Performing a weighted combination of the atmospheric refraction correction and the atmospheric scattering correction, dynamically adjusting the weight coefficient according to the time-varying characteristics of the corrected prediction data to obtain a comprehensive correction coefficient; and calculating the initial solar azimuth angle and initial solar altitude angle of the target photovoltaic power station area within a preset future time period based on the comprehensive correction coefficient;

[0023] The initial solar azimuth angle and the initial solar altitude angle are mutually constrained and calculated based on the geometric relationship between the solar azimuth angle and the solar altitude angle; the calculation results are corrected based on historical prediction errors to obtain the solar azimuth angle and the solar altitude angle of the target photovoltaic power station area within the future preset time period.

[0024] Establishing a motion constraint model for the photovoltaic tracking bracket based on the cloud coverage and wind speed data in the predicted meteorological data, determining whether the cloud coverage is greater than a preset constraint threshold and whether the wind speed data is greater than a safe wind speed threshold, and calculating a safe risk avoidance angle for the photovoltaic tracking bracket when the judgment conditions are met, and controlling the photovoltaic tracking bracket to rotate to the safe risk avoidance angle includes:

[0025] The motion constraint model includes wind load constraints and moment balance constraints; a trigger signal is generated based on the relationship between cloud cover and a preset constraint threshold, as well as the relationship between wind speed data and a safe wind speed threshold;

[0026] Obtaining structural parameters of a photovoltaic tracking bracket, the structural parameters including bracket area, bracket mass, current bracket inclination angle, and distance from the bracket center of gravity to the rotation axis; substituting the wind speed data, the bracket area, and the current bracket inclination angle into the wind load constraint to calculate a wind load value, and generating a wind torque based on the wind load value;

[0027] Calculate the gravity moment according to the mass of the bracket and the distance from the center of gravity of the bracket to the rotation axis; substitute the wind moment and the gravity moment into the moment balance constraint to calculate the safe avoidance angle that satisfies the motion constraint model;

[0028] Determine whether a risk avoidance control condition is met according to the trigger signal; when the risk avoidance control condition is met, obtain the current angle of the photovoltaic tracking bracket; calculate the target angular velocity based on the safe risk avoidance angle and the current angle, and control the photovoltaic tracking bracket to rotate to the safe risk avoidance angle according to the target angular velocity.

[0029] Based on the solar radiation intensity and temperature data in the predicted meteorological data, combined with the solar azimuth and the solar altitude angle, an improved dynamic programming algorithm is used to optimize and solve the optimal angle sequence of the photovoltaic tracking bracket within the future preset time period, including:

[0030] Acquire motion parameters of the photovoltaic tracking bracket, the motion parameters including a minimum rotation angle, a maximum rotation angle, and a rotation power consumption coefficient; establish an initial angle state space based on the minimum rotation angle and the maximum rotation angle;

[0031] Calculating an air quality index based on the solar azimuth angle and the solar altitude angle, and calculating direct radiation intensity based on the air quality index and solar radiation intensity; and multiplying the direct radiation intensity by the rotational power consumption coefficient to obtain a theoretical optimal power generation power;

[0032] Establishing a state density function based on the theoretical optimal power generation; adaptively dividing the initial angular state space according to the state density function, subdividing the region where the state density is higher than a preset density threshold, and obtaining an optimized angular state space; calculating the maximum power generation corresponding to each angular state in the optimized angular state space;

[0033] Calculating a temperature correction coefficient based on the temperature data and the temperature coefficient in the predicted meteorological data, multiplying the temperature correction coefficient by the maximum power generation to obtain the actual power generation; and constructing a power generation cost function by weighted combination of the actual power generation and the maximum power generation;

[0034] In the optimized angle state space, the power generation cost function is substituted into the dynamic programming recursive equation, and the optimal state value at each moment is obtained through iterative calculation; and the optimal angle sequence within a future preset time period is generated based on the optimal state value.

[0035] In the optimized angle state space, the power generation cost function is substituted into a dynamic programming recursive equation to obtain an optimal state value at each moment through iterative calculation; and generating an optimal angle sequence within a future preset time period based on the optimal state value includes:

[0036] Calculate the normalized difference between the actual generated power and the predicted generated power to obtain a power prediction error; substitute the power prediction error into an exponential decay function and multiply it by a basic discount rate to obtain a time-varying discount factor; add the negative value of the generated power cost function to the maximum value of the time-varying discount factor and the state value at the next moment to construct a dynamic programming recursive equation;

[0037] Calculating a current angle state value based on the dynamic programming recursive equation, performing weighted averaging of the current angle state value and the historical angle state value according to a preset learning rate to obtain an updated angle state value; calculating a relative error between the updated angle state value and the historical angle state value, and when the relative error is less than a preset convergence threshold, recording the updated angle state value as the final state value;

[0038] The angle that maximizes the final state value is selected from the optimized angle state space as the optimal angle at that moment; whether the optimal angle difference between adjacent predicted moments exceeds a preset maximum angle change rate is determined; if so, the optimal angle is corrected based on the maximum angle change rate; the corrected optimal angles are connected in sequence to generate an optimal angle sequence for the photovoltaic tracking bracket within a future preset time period.

[0039] Generating an angle adjustment instruction for the photovoltaic tracking bracket according to the optimal angle sequence includes:

[0040] The motion constraint parameters of the photovoltaic tracking bracket are obtained, wherein the motion constraint parameters include a maximum adjustment speed, a maximum acceleration, and a minimum adjustment time interval; the optimal angle sequence is subjected to trajectory smoothing based on the motion constraint parameters to generate a transition angle sequence that satisfies the maximum adjustment speed and the maximum acceleration; the transition angle sequence is subjected to time discretization sampling according to the minimum adjustment time interval, and the transition angle sequence after the discretization sampling is converted into an angle adjustment instruction.

[0041] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including:

[0042] processor;

[0043] a memory for storing processor-executable instructions;

[0044] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0045] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0046] The beneficial effects of this application are as follows:

[0047] By using a deep neural network model to predict future meteorological data and calculating the solar azimuth and altitude angles based on the predicted meteorological data, the photovoltaic tracking system's ability to perceive future lighting conditions is improved, making angle optimization forward-looking and avoiding the lag problem caused by traditional adjustments based on real-time data.

[0048] A motion constraint model is established based on cloud coverage and wind speed data. When severe weather conditions meet the constraint threshold, the system can automatically calculate the safe avoidance angle and control the rotation of the bracket, effectively ensuring the safe operation of photovoltaic equipment under extreme weather conditions and extending the service life of the equipment.

[0049] An improved dynamic programming algorithm is used to optimize and solve the optimal angle sequence, comprehensively considering solar radiation intensity and temperature factors. Compared with traditional fixed angle or simple tracking algorithms, it can maximize the power generation efficiency of photovoltaic modules under various meteorological conditions and improve the overall power generation and economic benefits of photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a method for optimizing the angle of photovoltaic tracking brackets and improving power generation efficiency based on weather forecasting according to an embodiment of the present invention;

[0051] Figure 2 This is a diagram showing the combined effect of the atmospheric refraction correction and the scattering correction according to an embodiment of the present invention;

[0052] Figure 3 This is a flowchart of the safety and risk avoidance control system for a photovoltaic tracking bracket according to an embodiment of the present invention;

[0053] Figure 4 This is a diagram showing the effect of the temperature correction coefficient on power generation efficiency according to an embodiment of the present invention;

[0054] Figure 5 This is a flow chart of the photovoltaic tracking control method based on dynamic programming optimization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0056] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0057] Figure 1 Schematic diagram of the process of optimizing the photovoltaic tracking bracket angle and improving the power generation efficiency for weather forecasting according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0058] Receiving meteorological data collected by a meteorological monitoring station for a target photovoltaic power station area, constructing a deep neural network model based on the meteorological data, and using the deep neural network model to predict meteorological data within a preset time period in the future to obtain predicted meteorological data;

[0059] Calculating the solar azimuth and solar altitude of the target photovoltaic power station area within the future preset time period based on the predicted meteorological data;

[0060] Based on the cloud coverage and wind speed data in the predicted meteorological data, a motion constraint model of the photovoltaic tracking bracket is established to determine whether the cloud coverage is greater than a preset constraint threshold and whether the wind speed data is greater than a safe wind speed threshold. When the judgment conditions are met, a safe risk avoidance angle of the photovoltaic tracking bracket is calculated, and the photovoltaic tracking bracket is controlled to rotate to the safe risk avoidance angle.

[0061] Based on the solar radiation intensity and temperature data in the predicted meteorological data, combined with the solar azimuth and the solar altitude angle, an improved dynamic programming algorithm is used to optimize and solve the optimal angle sequence of the photovoltaic tracking bracket within the future preset time period, wherein the optimal angle sequence is used to maximize the power generation efficiency of the photovoltaic assembly;

[0062] According to the optimal angle sequence, an angle adjustment instruction of the photovoltaic tracking bracket is generated, and the angle adjustment instruction is sent to a controller of the photovoltaic tracking bracket to control the photovoltaic tracking bracket to adjust the angle according to the optimal angle sequence.

[0063] In an optional embodiment, a deep neural network model is constructed based on the meteorological data, and the deep neural network model is used to predict meteorological data within a preset time period in the future, and the predicted meteorological data obtained includes:

[0064] Normalizing the meteorological data according to the adaptive time window length to obtain standardized features;

[0065] The deep neural network model includes a parallel feature extraction layer, an adaptive fusion layer, and a dynamic attention layer; the standardized features are input into the parallel feature extraction layer, and multi-scale local features are extracted through spatial pyramid convolution branches; the multi-scale local features are input into the adaptive fusion layer, adaptive weights are calculated based on feature discriminability, and feature fusion is performed to obtain enhanced features; the enhanced features are input into the dynamic attention layer, and key information at different time scales is captured based on a multi-scale temporal attention mechanism to obtain attention-enhanced features;

[0066] Constructing an adaptive multi-task loss function, the adaptive multi-task loss function including a prediction error loss term and a time series correlation loss term; adaptively adjusting the weight coefficients of each loss term based on the difficulty of the prediction task and the data distribution characteristics; and performing end-to-end training on the deep neural network model using the adaptive multi-task loss function;

[0067] The meteorological data to be predicted is input into the trained deep neural network model to generate predicted meteorological data.

[0068] Adaptive time window standardization is performed on meteorological data. In this embodiment, meteorological data includes multiple meteorological elements such as temperature, humidity, air pressure, wind speed and precipitation. After obtaining historical meteorological data, the system automatically determines the optimal time window length based on the data distribution characteristics and prediction task requirements. For example, for temperature data, the system determines that 72 hours is the optimal window length after analysis; while for precipitation data, a 48-hour window may be more suitable. After determining the window length, the data in each window is standardized, specifically including: calculating the mean and standard deviation of the data in the window, then subtracting the mean from the original data and dividing it by the standard deviation to obtain standardized features with a mean of 0 and a standard deviation of 1. Experiments show that compared with the standardization method with a fixed window length, the adaptive window can improve the prediction accuracy by approximately 5.8%.

[0069] A deep neural network model consisting of a parallel feature extraction layer, an adaptive fusion layer, and a dynamic attention layer was constructed. Within the parallel feature extraction layer, a spatial pyramid convolution branch was used to extract multi-scale local features. Specifically, three parallel convolution branches were set up with kernel sizes of 3×3, 5×5, and 7×7, respectively. Each branch consisted of two convolution layers, each followed by batch normalization and an activation function. For temperature prediction, for example, 72 hours of historical temperature data (of dimension 72×1) were input. After passing through the three parallel branches, feature maps of dimensions 70×32, 68×32, and 66×32 were generated, respectively. These feature maps captured local variation patterns within different receptive fields.

[0070] Adaptive weights are calculated based on feature discriminability and feature fusion is performed. Feature discriminability is measured by calculating the correlation between features and the target variable. Specifically, for each branch output feature, the correlation coefficient between it and the historical target variable is calculated. A higher correlation coefficient indicates that the feature is more important to the prediction and should be assigned a higher weight.

[0071] For example, in a prediction, the feature discriminants of the three branches are 0.65, 0.73, and 0.58, respectively. The corresponding adaptive weights are calculated as 0.33, 0.37, and 0.30. The features of the three branches are then weighted and summed to obtain the fused enhanced features. Experiments have shown that this adaptive fusion method can improve prediction accuracy by approximately 4.2% compared to simple feature concatenation or averaging.

[0072] A multi-scale temporal attention mechanism is used to capture key information at different time scales. Specifically, the enhanced features are fed into three attention units: short-term (6 hours), medium-term (24 hours), and long-term (72 hours). Each attention unit calculates the importance weight of each time point at the corresponding time scale, with the weight value ranging from 0 to 1, and ensures that the sum of all weights is 1.

[0073] For example, when predicting the temperature over the next 24 hours, the system might assign a total weight of 0.5 to the data from the last six hours, a weight of 0.3 to the data from the previous 24 hours, and a weight of 0.2 to earlier data. This allows the model to focus on the historical time period most relevant to the forecasting task, generating attention-enhancing features. Experiments have shown that the multi-scale temporal attention mechanism can improve the model's forecast accuracy for sudden weather changes by 7.6%.

[0074] An adaptive multi-task loss function is constructed to train the deep neural network model. This loss function includes a prediction error loss term and a time series correlation loss term. The prediction error loss term uses the mean absolute error to calculate the difference between the predicted value and the true value; the time series correlation loss term measures the consistency of the time trend between the predicted and true series.

[0075] The weight coefficients of each loss term are adaptively adjusted based on the difficulty of the forecast task and the characteristics of the data distribution. Specifically, for meteorological factors that fluctuate rapidly (such as precipitation), the system increases the weight of the time series correlation loss term; for relatively stable factors (such as air pressure), the weight of the prediction error loss term is increased. For example, in a temperature forecast task, the weight ratio of the prediction error loss term to the time series correlation loss term might be 0.7:0.3; while in precipitation forecasting, the weight ratio might be adjusted to 0.4:0.6.

[0076] The meteorological data to be predicted is input into the trained deep neural network model to generate the predicted meteorological data. In practical applications, the temperature prediction for the next 24 hours in Beijing's Haidian District is taken as an example: the historical temperature data for the last 72 hours is input, and after normalization through an adaptive time window, it passes through a parallel feature extraction layer, an adaptive fusion layer, and a dynamic attention layer, ultimately outputting hourly temperature forecasts for the next 24 hours. Experimental results show that the average absolute error of this method in temperature prediction is 0.7°C, which improves the prediction accuracy by 23.5% compared to traditional time series models. In the case of extreme weather changes, the prediction accuracy improves even more significantly, reaching 31.8%.

[0077] Through the above technical solution, the present invention can effectively capture the multi-scale characteristics and time dependence of meteorological data, achieve high-precision prediction of future meteorological data, and provide strong support for meteorological forecasting, agricultural production, disaster prevention and mitigation and other fields.

[0078] In an optional embodiment, calculating the solar azimuth angle and solar altitude angle of the target photovoltaic power station area within the future preset time period based on the predicted meteorological data includes:

[0079] Performing time series decomposition on the predicted meteorological data to extract time-varying features and periodic features; performing data correction on the predicted meteorological data based on the time-varying features and the periodic features to obtain corrected predicted data;

[0080] Calculate the atmospheric refraction correction value based on the predicted pressure data, predicted temperature data and the site altitude in the corrected predicted data; calculate the atmospheric scattering correction value based on the predicted relative humidity data, predicted cloud cover data and the site altitude in the corrected predicted data;

[0081] Performing a weighted combination of the atmospheric refraction correction and the atmospheric scattering correction, dynamically adjusting the weight coefficient according to the time-varying characteristics of the corrected prediction data to obtain a comprehensive correction coefficient; and calculating the initial solar azimuth angle and initial solar altitude angle of the target photovoltaic power station area within a preset future time period based on the comprehensive correction coefficient;

[0082] The initial solar azimuth angle and the initial solar altitude angle are mutually constrained and calculated based on the geometric relationship between the solar azimuth angle and the solar altitude angle; the calculation results are corrected based on historical prediction errors to obtain the solar azimuth angle and the solar altitude angle of the target photovoltaic power station area within the future preset time period.

[0083] The acquired forecasted meteorological data is decomposed into a time series to extract time-varying and cyclical features. Specifically, a seasonal trend decomposition method is used to decompose the forecasted meteorological data for the target PV power station area into three components: a trend term, a seasonal term, and a residual term. The trend term represents the long-term trend of the data and serves as a time-varying feature; the seasonal term represents the cyclical changes in the data and serves as a cyclical feature; and the residual term represents random fluctuations. For example, for a week's worth of temperature forecast data for a target PV power station area, decomposition yields a diurnal cyclical feature of gradually increasing daytime temperatures and decreasing nighttime temperatures, as well as a time-varying feature of gradually warming over the week.

[0084] The forecasted meteorological data is corrected based on the extracted time-varying and periodic features. Specifically, historical meteorological data for the same period is compared with the forecasted data, and a deviation is calculated. The forecasted data is then corrected based on the statistical distribution of the deviation. For example, if historical data shows that the forecast system's temperature predictions for the period from 2:00 PM to 4:00 PM are typically 2°C higher than the actual values, the forecasted temperature for that period is corrected by subtracting 2°C. This method results in more accurate forecasted data.

[0085] The atmospheric refraction correction is calculated based on the predicted pressure and temperature data in the corrected forecast data and the station altitude. Atmospheric refraction can cause the observed position of celestial objects to deviate from their actual positions. In practice, at an altitude of 500 meters, an air pressure of 101.3 kPa, and a temperature of 25°C, the calculated atmospheric refraction correction is approximately 0.08 degrees. This correction needs to be applied to the calculation of the solar altitude angle.

[0086] The atmospheric scattering correction is calculated based on the predicted relative humidity and cloud cover data in the corrected forecast data, as well as the station's elevation. Atmospheric scattering causes sunlight to scatter as it passes through the atmosphere, affecting the amount of direct solar radiation reaching the site. For example, when the relative humidity is 60%, the cloud cover is 30%, and the station's elevation is 200 meters, the calculated atmospheric scattering correction is approximately 0.12 degrees.

[0087] The atmospheric refraction correction and the atmospheric scattering correction are weighted together to form a comprehensive correction coefficient. The weighting coefficients are dynamically adjusted based on the time-varying characteristics of the corrected forecast data. For example, in clear weather, the refraction correction weight can be set to 0.6, and the scattering correction weight to 0.4; in cloudy weather, the refraction correction weight can be adjusted to 0.4, and the scattering correction weight to 0.6. This dynamic adjustment method produces a comprehensive correction coefficient that better reflects actual conditions.

[0088] The initial solar azimuth and altitude angles for the target PV power station area are calculated based on the comprehensive correction coefficients for a preset future time period. For example, for a PV power station located at 30 degrees north latitude and 120 degrees east longitude, at 12:00 PM on a certain day, the calculated initial solar azimuth and altitude angles are 178.5 degrees and 68.3 degrees, respectively.

[0089] The initial calculation results are mutually constrained based on the geometric relationship between the solar azimuth and solar altitude. There is a geometric constraint relationship between the solar azimuth and solar altitude, and it is necessary to ensure that the calculation results of the two are coordinated with each other. Specifically, through the joint inspection of parameters such as declination angle, hour angle, and geographic latitude, it is ensured that the calculation results of the two angles satisfy the spherical geometric relationship. When it is found that the constraint relationship is not satisfied, the other parameter is adjusted according to the parameter with higher confidence. For example, if the confidence of the solar altitude angle calculated based on geographic location and time is high, the azimuth angle is fine-tuned so that the two satisfy the geometric constraint relationship.

[0090] The calculation results are corrected based on historical prediction errors. By analyzing the sun angle prediction errors for the same location and similar weather conditions over a period of time, an error correction model is established. For example, if statistics show that the predicted sun angle in cloudy weather is typically 0.5 degrees lower than the actual value, this systematic error is compensated in the prediction results. In this way, the solar azimuth and solar altitude angles for the target PV power station area for a preset time period in the future are ultimately determined.

[0091] In a practical application at a photovoltaic power station, the above method was used to calculate the sun angle for the next 24 hours and then compared with the actual observed values. The results showed that compared with traditional methods, the average error in the calculated solar azimuth angle was reduced from 1.2 degrees to 0.5 degrees, and the average error in the solar altitude angle was reduced from 0.8 degrees to 0.3 degrees. This improved prediction accuracy by over 60%, effectively improving the power generation efficiency and power generation forecast accuracy of the photovoltaic power station.

[0092] Figure 2 This is a diagram showing the combined effect of the atmospheric refraction correction and the scattering correction according to an embodiment of the present invention:

[0093] This figure compares the performance of different photovoltaic power generation correction schemes over a full workday (6:00 AM - 6:00 PM). The figure includes four curves, representing the temporal trends of atmospheric refraction correction (circular lines), atmospheric scattering correction (triangular lines), the combined correction coefficient of this technical scheme (square lines), and the single correction coefficient of the traditional method (cross-shaped lines). The data shows that all correction coefficients exhibit a U-shaped distribution, reaching their minimum value at noon and increasing in the morning and evening. Specifically, the combined correction coefficient of this technical scheme is approximately 0.62 at 6:00 AM, then gradually decreases, reaching a minimum of approximately 0.21 at 12:00 PM, then begins to recover, returning to approximately 0.65 by 6:00 PM. The atmospheric refraction correction ranges from 0.15 to 0.42, reaching a minimum of approximately 0.15 at noon. The atmospheric scattering correction is relatively small, ranging from 0.09 to 0.29, with a maximum of approximately 0.09 at noon. The overall performance of the traditional method's single correction coefficient lies between atmospheric refraction and the combined correction of our solution, reaching approximately 0.50 at 6:00 AM, dropping to 0.21 at noon, and rising to 0.52 at 6:00 PM. This comparison demonstrates that our solution, by considering the combined effects of atmospheric refraction and scattering, provides a more comprehensive correction capability, demonstrating a clear advantage during critical periods such as sunrise and sunset.

[0094] In an optional embodiment, a motion constraint model of a photovoltaic tracking bracket is established based on the cloud coverage and wind speed data in the predicted meteorological data, and it is determined whether the cloud coverage is greater than a preset constraint threshold and whether the wind speed data is greater than a safe wind speed threshold. When the judgment conditions are met, a safe risk avoidance angle of the photovoltaic tracking bracket is calculated, and controlling the photovoltaic tracking bracket to rotate to the safe risk avoidance angle includes:

[0095] The motion constraint model includes wind load constraints and moment balance constraints; a trigger signal is generated based on the relationship between cloud cover and a preset constraint threshold, as well as the relationship between wind speed data and a safe wind speed threshold;

[0096] Obtaining structural parameters of a photovoltaic tracking bracket, the structural parameters including bracket area, bracket mass, current bracket inclination angle, and distance from the bracket center of gravity to the rotation axis; substituting the wind speed data, the bracket area, and the current bracket inclination angle into the wind load constraint to calculate a wind load value, and generating a wind torque based on the wind load value;

[0097] Calculate the gravity moment according to the mass of the bracket and the distance from the center of gravity of the bracket to the rotation axis; substitute the wind moment and the gravity moment into the moment balance constraint to calculate the safe avoidance angle that satisfies the motion constraint model;

[0098] Determine whether a risk avoidance control condition is met according to the trigger signal; when the risk avoidance control condition is met, obtain the current angle of the photovoltaic tracking bracket; calculate the target angular velocity based on the safe risk avoidance angle and the current angle, and control the photovoltaic tracking bracket to rotate to the safe risk avoidance angle according to the target angular velocity.

[0099] Obtain cloud cover and wind speed data from forecasted weather data. Weather data can be obtained in real time from weather service platforms. For example, you can obtain 24-hour forecast data from the Meteorological Bureau or a third-party weather service provider through an API, including hourly cloud cover (as a percentage) and wind speed data (meters per second).

[0100] A motion constraint model of the photovoltaic tracking bracket is established. The model includes two parts: wind load constraint and moment balance constraint.

[0101] Wind load constraints primarily consider the force exerted by wind on the PV mount. The wind load value is proportional to the square of the wind speed, the mount area, and the sine of the mount inclination angle. Specifically, the wind load value is equal to the square of the wind speed multiplied by the mount area multiplied by the sine of the inclination angle, multiplied by the drag coefficient (typically 0.6) multiplied by half the air density (typically 1.225 kg / m³). For example, when the wind speed is 15 m / s, the mount area is 20 m², and the current mount inclination angle is 30 degrees, the calculated wind load value is approximately 1378.1 Newtons.

[0102] The moment balance constraint considers the balance between wind moment and gravity moment. The wind moment is equal to the wind load multiplied by the vertical distance from the center of the support to the axis of rotation. The gravity moment is equal to the support mass multiplied by the acceleration due to gravity multiplied by the horizontal distance from the support's center of gravity to the axis of rotation. The safe avoidance angle should ensure that the wind moment is less than or equal to the gravity moment to ensure system stability.

[0103] Determine whether the hazard avoidance conditions are met. Set a preset constraint threshold (e.g., 80% cloud cover) and a safe wind speed threshold (e.g., 10 m / s). When the cloud cover exceeds the preset constraint threshold and the wind speed exceeds the safe wind speed threshold, a trigger signal is generated, indicating that hazard avoidance control is required.

[0104] Obtain the structural parameters of the photovoltaic tracking bracket, including the bracket area (for example, 20 square meters), bracket mass (for example, 500 kilograms), current bracket inclination angle (for example, 35 degrees), and the distance from the bracket center of gravity to the rotation axis (for example, 0.8 meters).

[0105] Substitute the wind speed data, support area, and current support inclination angle into the wind load constraint. Assuming the current wind speed is 12 m / s, the support area is 20 m2, and the current support inclination angle is 35 degrees, substituting these into the wind load constraint formula yields a wind load value of approximately 982.7 Newtons.

[0106] The wind moment is generated based on the wind load value. Assuming the vertical distance from the center of the bracket to the axis of rotation is 0.5 meters, the wind moment is approximately 491.4 Newton meters.

[0107] Calculate the gravity moment based on the mass of the bracket and the distance from the bracket's center of gravity to the axis of rotation. Assuming the bracket mass is 500 kg and the horizontal distance from the bracket's center of gravity to the axis of rotation is 0.8 meters multiplied by the cosine of the current inclination angle, that is:

[0108] 0.8×cos(35°)=0.655 m, so the gravitational moment is equal to 500×9.8×0.655=3209.5 Newton meters.

[0109] Substituting the wind torque and gravity torque into the torque balance constraints, the safe avoidance angle that satisfies the motion constraint model is calculated. In this example, as the bracket inclination angle increases, the wind torque decreases, and the gravity torque also decreases. By adjusting the bracket inclination angle, the maximum angle at which the wind torque is less than the gravity torque is the safe avoidance angle. Calculations show that when the bracket angle is adjusted to 75 degrees, the wind torque decreases to approximately 260 Newton-meters and the gravity torque is 1270 Newton-meters. At this point, the system is stable, and 75 degrees can be determined as the safe avoidance angle.

[0110] The trigger signal is used to determine whether the risk avoidance control condition is met. If the risk avoidance control condition is met, the current angle of the photovoltaic tracking bracket is obtained, assuming that the current angle is 35 degrees.

[0111] The target angular velocity is calculated based on the safe avoidance angle (75 degrees) and the current angle (35 degrees). The angle difference is 40 degrees. If the adjustment time is set to 60 seconds, the target angular velocity is 40 degrees / 60 seconds = 0.67 degrees / second.

[0112] The PV tracking mount is controlled to rotate to a safe angle based on the target angular velocity. The motor speed is controlled to rotate at a rate of 0.67 degrees per second. After approximately 60 seconds, the mount reaches a safe angle of 75 degrees and maintains this position until weather conditions improve.

[0113] In actual applications, to ensure system stability and reliability, you can set multiple wind speed thresholds and corresponding safety angles. For example, when the wind speed is 10-15 m / s, the safety angle is 60 degrees; when the wind speed is 15-20 m / s, the safety angle is 75 degrees; when the wind speed exceeds 20 m / s, the safety angle is 90 degrees (fully vertical).

[0114] Furthermore, the preset constraint thresholds and safe wind speed thresholds can be dynamically adjusted based on the meteorological characteristics of different regions and seasons to adapt to operational needs under different environmental conditions. For example, in coastal areas prone to typhoons, the safe wind speed threshold can be set lower to facilitate risk avoidance control in advance.

[0115] The method of the present invention can effectively prevent damage to photovoltaic tracking brackets caused by severe weather, extend the service life of the equipment, and improve the safety and reliability of the photovoltaic power generation system.

[0116] Figure 3 The following is a flowchart of the safety and risk avoidance control system for photovoltaic tracking brackets according to an embodiment of the present invention:

[0117] This is a flow chart for the safety and avoidance control of a photovoltaic tracker, detailing the complete control process from meteorological data acquisition to final avoidance action. First, the system acquires predicted meteorological data, including cloud cover and wind speed data, and compares them with preset thresholds. If the cloud cover exceeds the preset threshold or the wind speed exceeds the safe wind speed threshold, the avoidance control process is triggered. If it does not exceed the threshold, the system maintains normal tracking mode, without triggering avoidance control. After entering the avoidance control process, the system first establishes a motion constraint model, including two key constraints: wind load constraint and torque balance constraint. It then acquires the structural parameters of the photovoltaic tracker, including core parameters such as the tracker area, mass, inclination angle, and the distance from the center of gravity to the rotation axis. Based on these parameters, the system calculates wind load and gravity moment. The wind load calculation primarily considers the impact of wind torque on the tracker, while the gravity moment calculation is based on the tracker mass and structural parameters. These two calculation results are substituted into the torque balance constraint equation to solve for the safe avoidance angle. Finally, during the control execution phase, the system acquires the current angle, calculates the target angular velocity, and controls the tracker to rotate to the safe avoidance angle. This process ensures the safe operation of photovoltaic tracking brackets in severe weather conditions through rigorous mechanical analysis and real-time control.

[0118] In an optional embodiment, based on the solar radiation intensity and temperature data in the predicted meteorological data, combined with the solar azimuth and the solar altitude angle, an improved dynamic programming algorithm is used to optimize and solve the optimal angle sequence of the photovoltaic tracking bracket within the future preset time period, including:

[0119] Acquire motion parameters of the photovoltaic tracking bracket, the motion parameters including a minimum rotation angle, a maximum rotation angle, and a rotation power consumption coefficient; establish an initial angle state space based on the minimum rotation angle and the maximum rotation angle;

[0120] Calculating an air quality index based on the solar azimuth angle and the solar altitude angle, and calculating direct radiation intensity based on the air quality index and solar radiation intensity; and multiplying the direct radiation intensity by the rotational power consumption coefficient to obtain a theoretical optimal power generation power;

[0121] Establishing a state density function based on the theoretical optimal power generation; adaptively dividing the initial angular state space according to the state density function, subdividing the region where the state density is higher than a preset density threshold, and obtaining an optimized angular state space; calculating the maximum power generation corresponding to each angular state in the optimized angular state space;

[0122] Calculating a temperature correction coefficient based on the temperature data and the temperature coefficient in the predicted meteorological data, multiplying the temperature correction coefficient by the maximum power generation to obtain the actual power generation; and constructing a power generation cost function by weighted combination of the actual power generation and the maximum power generation;

[0123] In the optimized angle state space, the power generation cost function is substituted into the dynamic programming recursive equation, and the optimal state value at each moment is obtained through iterative calculation; and the optimal angle sequence within a future preset time period is generated based on the optimal state value.

[0124] Obtain the motion parameters of the photovoltaic tracker, including the minimum rotation angle, maximum rotation angle, and rotation power consumption coefficient. In this embodiment, the minimum rotation angle of the photovoltaic tracker is -60 degrees, the maximum rotation angle is 60 degrees, and the rotation power consumption coefficient is 0.02 kWh / degree. Based on these parameters, establish an initial angular state space, i.e., an angular range from -60 degrees to 60 degrees.

[0125] The AMI is calculated based on the solar azimuth and solar altitude. In practice, when the solar altitude is 45 degrees, the AMI is approximately 1.4; when the solar altitude drops to 30 degrees, the AMI increases to approximately 2.0. For example, at 10 a.m. on a given day, assuming the solar altitude is 35 degrees and the azimuth is 135 degrees, the AMI is calculated to be 1.7.

[0126] Direct radiation intensity is calculated based on the Air Quality Index and predicted solar radiation intensity. For example, when the predicted total radiation intensity is 800 watts per square meter and the Air Quality Index is 1.7, the direct radiation intensity is approximately 650 watts per square meter. Multiplying this direct radiation intensity by the rotational power consumption coefficient of 0.02 kWh / kWh yields a theoretical optimal power generation capacity of 13 kilowatts.

[0127] A state density function is established based on the theoretical optimal power generation. This function reflects the distribution of power generation at different angles. In this embodiment, when the PV mount angle is close to the solar incident angle, the state density value is high; when the angle deviates significantly, the state density value decreases rapidly.

[0128] The initial angular state space is adaptively partitioned based on the state density function. A preset density threshold of 0.8 is set, and regions with state density above this threshold are subdivided. For example, within the angular range of 25 to 40 degrees, the state density is generally above 0.8, so this region is subdivided into 1-degree intervals. In the range of -60 to -20 degrees, the state density is lower, so 5-degree intervals are maintained. This method results in an optimized angular state space that ensures computational accuracy in high-density regions while reducing computational effort.

[0129] After optimizing the angle state space, the maximum power generation corresponding to each angle state in the optimized angle state space is calculated. For example, at a specific moment, when the bracket angle is 35 degrees, considering factors such as the incident angle and the area of the photovoltaic panel, the calculated maximum power generation is 12.5 kilowatts. At 20 degrees, the maximum power generation drops to 11.2 kilowatts, and at -30 degrees, the maximum power generation is only 5.6 kilowatts.

[0130] The temperature correction factor is calculated based on the temperature data and the temperature coefficient from the forecasted meteorological data. Assuming a PV panel temperature coefficient of -0.4% / °C, when the ambient temperature is 30°C and the panel temperature rises to 45°C, the temperature rise is 20°C compared to the standard test conditions of 25°C. The temperature correction factor is 1-0.4% × 20 = 0.92. This temperature correction factor is multiplied by the maximum power generation to obtain the actual power generation. For example, at an angle of 35 degrees, the actual power generation is 12.5 × 0.92 = 11.5 kilowatts.

[0131] The power generation cost function is constructed by weighting the actual power generation and the maximum power generation. In this embodiment, the weight coefficients are 0.7 and 0.3 respectively, that is, the cost function value = 0.7 × actual power generation + 0.3 × maximum power generation. For example, when the angle is 35 degrees, the cost function value is:

[0132] 0.7×11.5+0.3×12.5=11.8 kilowatts.

[0133] In the optimized angular state space, the power generation cost function is substituted into the dynamic programming recursive equation, and the optimal state value at each moment is obtained through iterative calculation. Assuming a preset time period of one day, with one-hour intervals, a total of 24 time points, the optimal angular state at each time point is determined through recursive calculation.

[0134] Generates the optimal angle sequence for a preset future time period based on the optimal state values. For example, the optimal angle sequence from 6:00 AM to 6:00 PM on a certain day is: -45 degrees (6:00), -35 degrees (7:00), -25 degrees (8:00), -10 degrees (9:00), 5 degrees (10:00), 18 degrees (11:00), 30 degrees (12:00), 35 degrees (13:00), 32 degrees (14:00), 25 degrees (15:00), 10 degrees (16:00), -5 degrees (17:00), -20 degrees (18:00).

[0135] Through the above steps, the method of the present invention comprehensively considers the changes in the sun's position, predicted meteorological data, temperature effects, and the energy consumption of the bracket's rotation. Using an improved dynamic programming algorithm, it achieves optimal control of the photovoltaic tracking bracket's angle. Compared to traditional methods, this method increased power generation by 8.5% in actual tests, demonstrating its excellent practical value.

[0136] Figure 4 This is a diagram showing the effect of the temperature correction coefficient on power generation efficiency according to an embodiment of the present invention:

[0137] This figure shows the changing trends of various PV system performance parameters over a full workday (8:00 AM - 5:00 PM). The figure includes four curves, representing the temporal relationships between ambient temperature (cross-shaped line), temperature correction coefficient (triangle line), power generation efficiency after correction using this solution (circle line), and the uncorrected power generation efficiency using a conventional solution (square line). The data shows that the ambient temperature gradually rises from around 22°C at 8:00 AM, reaching a daily peak of approximately 36°C at 1:00 PM, before beginning to decline. The temperature correction coefficient exhibits an inverse trend to the ambient temperature, gradually decreasing from around 0.97 in the morning to around 0.90 at noon. The power generation efficiency after correction using this solution remains high overall, starting at 92% at 8:00 AM, peaking at around 95% at 10:00 AM, and remaining above 92% even during the hottest afternoon hours, before significantly declining to around 86% after 4:00 PM. In contrast, the traditional solution's uncorrected power generation efficiency performed poorly overall, starting at 86% in the morning. Although it saw a slight increase to 88% between 9:00 and 10:00, it dropped significantly during the hot midday hours, reaching a low of approximately 82% at 1:00 PM. Even after the temperature dropped in the afternoon, it only recovered to around 84%. This demonstrates that the proposed technology can effectively improve the system's power generation efficiency through temperature correction, especially during hot periods.

[0138] In an optional embodiment, in the optimized angle state space, substituting the power generation cost function into a dynamic programming recursive equation to obtain an optimal state value at each moment through iterative calculation; and generating an optimal angle sequence within a future preset time period based on the optimal state value includes:

[0139] Calculate the normalized difference between the actual generated power and the predicted generated power to obtain a power prediction error; substitute the power prediction error into an exponential decay function and multiply it by a basic discount rate to obtain a time-varying discount factor; add the negative value of the generated power cost function to the maximum value of the time-varying discount factor and the state value at the next moment to construct a dynamic programming recursive equation;

[0140] Calculating a current angle state value based on the dynamic programming recursive equation, performing weighted averaging of the current angle state value and the historical angle state value according to a preset learning rate to obtain an updated angle state value; calculating a relative error between the updated angle state value and the historical angle state value, and when the relative error is less than a preset convergence threshold, recording the updated angle state value as the final state value;

[0141] The angle that maximizes the final state value is selected from the optimized angle state space as the optimal angle at that moment; whether the optimal angle difference between adjacent predicted moments exceeds a preset maximum angle change rate is determined; if so, the optimal angle is corrected based on the maximum angle change rate; the corrected optimal angles are connected in sequence to generate an optimal angle sequence for the photovoltaic tracking bracket within a future preset time period.

[0142] In the optimized angle state space, the power generation cost function is substituted into the dynamic programming recursive equation, and the optimal state value at each moment is obtained through iterative calculation. Based on the optimal state value, the optimal angle sequence for a preset future time period is generated.

[0143] The initial angular state space of a photovoltaic system can be set to a set of angles in 5-degree intervals from 0° to 90°, namely {0°, 5°, 10°, 15°, ..., 90°}. After considering actual lighting conditions, the angular state space can be optimized to a smaller range. For example, the angle state space might be optimized to {30°, 35°, 40°, ..., 60°} for the period from 8:00 AM to 4:00 PM to improve computational efficiency.

[0144] The first step in the dynamic programming solution is to calculate the normalized difference between the actual and predicted power generation to obtain the power forecast error. Assuming the current actual power generation is 580W and the predicted power generation is 600W, the normalized difference is calculated as (580-600) / 600 = -0.033, which means the power forecast error is -3.3%.

[0145] Substituting the power forecast error into the exponential decay function and multiplying it by the base discount rate yields the time-varying discount factor. Assuming the exponential decay function is a negative exponential, the base discount rate is 0.95, and the decay coefficient is 2, the time-varying discount factor is calculated as 0.95 × (1 - 0.033)² ≈ 0.944, representing the weight coefficient of the impact of current decisions on the future.

[0146] Construct a dynamic programming recursive equation, adding the negative value of the power cost function to the time-varying discount factor and the maximum state value at the next moment. For example, when the power cost function value at an angle of 45° is -590W, the time-varying discount factor is 0.944, and the maximum state value at the next moment is 620, then the current recursive value is -590 + 0.944 × 620 ≈ -5.72.

[0147] The current angle state value is calculated using a dynamic programming recursive equation. The updated angle state value is obtained by taking a weighted average of the current angle state value and the historical angle state value according to the preset learning rate. Assuming the current calculated state value is 595, the historical state value is 590, and the learning rate is 0.2, the updated state value is 590 + 0.2 × (595 - 590) = 591, which represents the comprehensive evaluation value when the angle is 45°.

[0148] Calculate the relative error between the updated angle state value and the historical angle state value. The relative error is calculated as |(591-590) / 590|=0.0017. If the preset convergence threshold is set to 0.001, the relative error is greater than the threshold at this point, and further iteration is required. Suppose that after multiple iterations, the updated state value is 590.5 and the historical state value is 590.4. The relative error is |(590.5-590.4) / 590.4|≈0.00017<0.001. At this point, convergence conditions are met, and 590.5 is recorded as the final state value of the angle.

[0149] Repeat the above steps for all angles in the optimized angular state space. For example, the final state values calculated for different angles are: 40° corresponds to 580.2, 45° corresponds to 590.5, and 50° corresponds to 585.6. The angle 45° that maximizes the final state value is selected as the optimal angle at the current moment.

[0150] Determine whether the difference in optimal angles between adjacent prediction times exceeds the preset maximum angle change rate. Assume the optimal angles at two adjacent times are 45° and 60°, respectively, with a 10-minute interval and a preset maximum angle change rate of 1° / minute. The 15° angle difference exceeds the maximum allowable change of 10° and requires correction. The corrected angle at the next moment should be 45° + 10° = 55°, rather than the original 60°.

[0151] All corrected optimal angles are sequentially connected to generate the optimal angle sequence for the PV tracker within a preset future time period. For example, from 8:00 AM to 4:00 PM on a given day, the optimal angle sequence for each half-hour might be {35°, 40°, 45°, 50°, 55°, 60°, 55°, 50°, 45°, 40°, 35°, 30°, 25°}. This sequence takes into account the changing trajectory of the sun and system constraints, maximizing the PV system's power generation.

[0152] In practical applications, various parameters can be adjusted based on specific system parameters. For example, the calculation of power prediction error can use a sliding window average to reduce the impact of instantaneous fluctuations; the attenuation coefficient of the time-varying discount factor can be adjusted according to weather conditions, with a setting of 1.5 on sunny days and 2.5 on cloudy days; the learning rate can be adjusted based on system response speed, with a setting of 0.1 for large systems and 0.3 for small systems; the convergence threshold can be set based on accuracy requirements, generally 0.0005 for high-precision applications and 0.002 for general applications; and the maximum angle change rate can be set based on mechanical performance, with a setting of 2° / minute for high-performance drives and 1° / minute for general drives.

[0153] This method effectively solves the optimization problem of photovoltaic tracking control through a dynamic programming algorithm, and generates the optimal angle sequence for the photovoltaic tracking bracket while considering the real-time power generation power, prediction error and system constraints, thereby improving the overall power generation efficiency of the system.

[0154] Figure 5 Flowchart of the photovoltaic tracking control method for dynamic programming optimization according to an embodiment of the present invention:

[0155] The power prediction error is calculated by calculating the normalized difference between the actual and predicted power generation. This error is then substituted into an exponential decay function and multiplied by a base discount rate to obtain a time-varying discount factor. The negative value of the power cost function is then added to the time-varying discount factor and the maximum state value at the next moment to construct a dynamic programming recursive equation, laying the foundation for subsequent optimization. Based on this constructed dynamic programming recursive equation, the current angle state value is calculated and weighted averaged with the historical angle state values according to a pre-set learning rate to obtain the updated angle state value. The relative error between the updated and historical angle state values is then calculated. When this error is less than a preset convergence threshold, the updated angle state value is recorded as the final state value. For each prediction moment in the optimization space, the angle that maximizes the final state value is selected as the optimal angle at that moment. A constraint is imposed by determining whether the difference between the optimal angles at adjacent prediction moments exceeds a preset maximum angle change rate. If this exceeds the threshold, corrections are made based on the maximum angle change rate. Finally, the corrected optimal angles at each moment are sequentially connected to generate a complete optimal angle control sequence.

[0156] In an optional embodiment, generating an angle adjustment instruction for the photovoltaic tracking bracket according to the optimal angle sequence includes:

[0157] The motion constraint parameters of the photovoltaic tracking bracket are obtained, wherein the motion constraint parameters include a maximum adjustment speed, a maximum acceleration, and a minimum adjustment time interval; the optimal angle sequence is subjected to trajectory smoothing based on the motion constraint parameters to generate a transition angle sequence that satisfies the maximum adjustment speed and the maximum acceleration; the transition angle sequence is subjected to time discretization sampling according to the minimum adjustment time interval, and the transition angle sequence after the discretization sampling is converted into an angle adjustment instruction.

[0158] Obtaining the motion constraint parameters of the photovoltaic tracking bracket is the basic step in generating angle adjustment instructions. The motion constraint parameters mainly include the maximum adjustment speed, maximum acceleration and minimum adjustment time interval. The maximum adjustment speed refers to the maximum angular velocity allowed by the photovoltaic tracking bracket during the adjustment process. It is usually determined based on the characteristics of the bracket's drive motor and the strength of the mechanical structure. In practical applications, the typical value is set to 1.5 degrees / second. The maximum acceleration refers to the maximum angular acceleration allowed by the bracket during the angle adjustment process. This parameter is related to the smoothness of the bracket adjustment process and the mechanical stress condition. The typical value is set to 0.3 degrees / second. 2 The minimum adjustment interval is the minimum time interval between two angle adjustment commands. This parameter takes into account the control system's processing power and execution efficiency, and is typically set to 5 seconds. These parameters can be read from the support system configuration file and can also be dynamically adjusted based on the actual operating environment.

[0159] Based on the motion constraint parameters, the trajectory of the optimal angle sequence is smoothed to generate a transition angle sequence that meets the maximum adjustment speed and maximum acceleration. The trajectory smoothing method uses the cubic spline interpolation method to construct a continuous curve between adjacent angle points to ensure that the angle change transitions smoothly within the speed and acceleration constraints. In specific implementation, the angle difference between adjacent angle points in the optimal angle sequence is first calculated, and then the minimum transition time is determined based on the maximum adjustment speed. When the calculated transition time is less than the minimum time calculated based on the acceleration constraint, the acceleration-uniform speed-deceleration adjustment mode is adopted; when the calculated transition time is greater than the minimum time calculated based on the acceleration constraint, the acceleration-deceleration mode is directly adopted without going through the uniform speed stage.

[0160] Assume that the two adjacent angle points in the optimal angle sequence are 30 degrees and 45 degrees respectively, and the angle difference is 15 degrees. According to the maximum adjustment speed of 1.5 degrees / second, it takes at least 10 seconds to complete the adjustment; and according to the maximum acceleration of 0.3 degrees / second 2Calculation shows that it takes 5 seconds to accelerate to the maximum speed and 5 seconds to decelerate, for a total of 10 seconds. In this case, the acceleration-deceleration mode is used, and no uniform speed segment is required. The transition angle sequence obtained after trajectory smoothing is a set of angle values that smoothly transition from 30 degrees to 45 degrees within 10 seconds, ensuring that the angular velocity does not exceed 1.5 degrees / second and the angular acceleration does not exceed 0.3 degrees / second during the entire process. 2 .

[0161] The transition angle sequence is sampled in a time-discrete manner according to the minimum adjustment time interval, and the discretized transition angle sequence is converted into angle adjustment instructions. Time-discrete sampling involves sampling the continuous transition angle sequence at time intervals that the control system can handle, converting it into a discrete sequence of angle points. The sampling interval must meet the minimum adjustment time interval requirements while also considering the real-time requirements of the control system. In practical applications, if the minimum adjustment time interval is 5 seconds, an angle value is extracted every 5 seconds from the smoothed continuous transition angle sequence to form a discrete angle sequence.

[0162] Converting the discretely sampled transition angle sequence into angle adjustment commands involves two aspects: command format construction and command queue management. Command format construction encapsulates the angle values into a command format recognizable by the control system, typically including information such as timestamp, target angle, and execution priority. Command queue management is responsible for storing the constructed commands in the command queue in execution order for the control system to execute them in sequence.

[0163] Assuming a minimum adjustment interval of 5 seconds, three angle adjustment points are obtained through discretized sampling: 30 degrees at 0 seconds, 37.5 degrees at 5 seconds, and 45 degrees at 10 seconds. These angle values are converted into angle adjustment commands, forming the following command sequence: [Timestamp_1, Angle: 30 degrees, Priority: Normal], [Timestamp_2, Angle: 37.5 degrees, Priority: Normal], [Timestamp_3, Angle: 45 degrees, Priority: Normal]. These commands are sequentially sent to the actuator, driving the photovoltaic tracker to adjust its angle according to the predetermined trajectory.

[0164] In more complex application scenarios, the optimal angle sequence may include multiple angle points, and the intervals between adjacent angle points may be very small. For example, the optimal angle sequence for a photovoltaic power station in a day includes 48 angle points from east to west, with an average of one angle point every 30 minutes. Generating adjustment instructions directly based on these angle points may lead to frequent bracket adjustments, increasing energy consumption and shortening equipment life. Through trajectory smoothing and time discretization sampling, these 48 angle points can be converted into a smooth transition sequence that meets motion constraints, reducing unnecessary adjustments and optimizing the bracket's motion trajectory.

[0165] In practical applications, this method significantly improves the efficiency and smoothness of angular adjustment for photovoltaic trackers. By acquiring motion constraint parameters in real time, dynamically calculating transition angle sequences, and performing time-discrete sampling, the photovoltaic tracker can move along an optimal trajectory, reducing mechanical stress, extending the equipment life, and ensuring the efficient operation of the photovoltaic power generation system.

[0166] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including:

[0167] processor;

[0168] a memory for storing processor-executable instructions;

[0169] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0170] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0171] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the angle of photovoltaic tracking brackets and improving power generation efficiency based on weather forecasting, characterized in that: include: Receiving meteorological data collected by a meteorological monitoring station for a target photovoltaic power station area, building a deep neural network model based on the meteorological data, and using the deep neural network model to predict meteorological data within a preset time period in the future to obtain predicted meteorological data, including: Normalizing the meteorological data according to the adaptive time window length to obtain standardized features; The deep neural network model includes a parallel feature extraction layer, an adaptive fusion layer, and a dynamic attention layer; the standardized features are input into the parallel feature extraction layer, and multi-scale local features are extracted through spatial pyramid convolution branches; the multi-scale local features are input into the adaptive fusion layer, adaptive weights are calculated based on feature discriminability, and feature fusion is performed to obtain enhanced features; the enhanced features are input into the dynamic attention layer, and key information at different time scales is captured based on a multi-scale temporal attention mechanism to obtain attention-enhanced features; Constructing an adaptive multi-task loss function, the adaptive multi-task loss function including a prediction error loss term and a time series correlation loss term; adaptively adjusting the weight coefficients of each loss term based on the difficulty of the prediction task and the data distribution characteristics; and performing end-to-end training on the deep neural network model using the adaptive multi-task loss function; Input the meteorological data to be predicted into the trained deep neural network model to generate predicted meteorological data; Calculating the solar azimuth and solar altitude of the target photovoltaic power station area within the future preset time period based on the predicted meteorological data; Based on the cloud coverage and wind speed data in the predicted meteorological data, a motion constraint model of the photovoltaic tracking bracket is established to determine whether the cloud coverage is greater than a preset constraint threshold and whether the wind speed data is greater than a safe wind speed threshold. When the judgment conditions are met, a safe risk avoidance angle of the photovoltaic tracking bracket is calculated, and the photovoltaic tracking bracket is controlled to rotate to the safe risk avoidance angle. Based on the solar radiation intensity and temperature data in the predicted meteorological data, combined with the solar azimuth and the solar altitude angle, an improved dynamic programming algorithm is used to optimize and solve the optimal angle sequence of the photovoltaic tracking bracket within the future preset time period, wherein the optimal angle sequence is used to maximize the power generation efficiency of the photovoltaic assembly; According to the optimal angle sequence, an angle adjustment instruction of the photovoltaic tracking bracket is generated, and the angle adjustment instruction is sent to a controller of the photovoltaic tracking bracket to control the photovoltaic tracking bracket to adjust the angle according to the optimal angle sequence.

2. The method according to claim 1, characterized in that Calculating the solar azimuth and solar altitude angle of the target photovoltaic power station area within the future preset time period based on the predicted meteorological data includes: Performing time series decomposition on the predicted meteorological data to extract time-varying features and periodic features; performing data correction on the predicted meteorological data based on the time-varying features and the periodic features to obtain corrected predicted data; Calculate the atmospheric refraction correction value based on the predicted pressure data, predicted temperature data and the site altitude in the corrected predicted data; calculate the atmospheric scattering correction value based on the predicted relative humidity data, predicted cloud cover data and the site altitude in the corrected predicted data; Performing a weighted combination of the atmospheric refraction correction and the atmospheric scattering correction, dynamically adjusting the weight coefficient according to the time-varying characteristics of the corrected prediction data to obtain a comprehensive correction coefficient; and calculating the initial solar azimuth angle and initial solar altitude angle of the target photovoltaic power station area within a preset future time period based on the comprehensive correction coefficient; The initial solar azimuth angle and the initial solar altitude angle are mutually constrained and calculated based on the geometric relationship between the solar azimuth angle and the solar altitude angle; the calculation results are corrected based on historical prediction errors to obtain the solar azimuth angle and the solar altitude angle of the target photovoltaic power station area within the future preset time period.

3. The method according to claim 1, characterized in that Establishing a motion constraint model for the photovoltaic tracking bracket based on the cloud coverage and wind speed data in the predicted meteorological data, determining whether the cloud coverage is greater than a preset constraint threshold and whether the wind speed data is greater than a safe wind speed threshold, and calculating a safe risk avoidance angle for the photovoltaic tracking bracket when the judgment conditions are met, and controlling the photovoltaic tracking bracket to rotate to the safe risk avoidance angle includes: The motion constraint model includes wind load constraints and moment balance constraints; a trigger signal is generated based on the relationship between cloud cover and a preset constraint threshold, as well as the relationship between wind speed data and a safe wind speed threshold; Obtaining structural parameters of a photovoltaic tracking bracket, the structural parameters including bracket area, bracket mass, current bracket inclination angle, and distance from the bracket center of gravity to the rotation axis; substituting the wind speed data, the bracket area, and the current bracket inclination angle into the wind load constraint to calculate a wind load value, and generating a wind torque based on the wind load value; Calculate the gravity moment according to the mass of the bracket and the distance from the center of gravity of the bracket to the rotation axis; substitute the wind moment and the gravity moment into the moment balance constraint to calculate the safe avoidance angle that satisfies the motion constraint model; Determine whether a risk avoidance control condition is met according to the trigger signal; when the risk avoidance control condition is met, obtain the current angle of the photovoltaic tracking bracket; calculate the target angular velocity based on the safe risk avoidance angle and the current angle, and control the photovoltaic tracking bracket to rotate to the safe risk avoidance angle according to the target angular velocity.

4. The method according to claim 1, wherein Based on the solar radiation intensity and temperature data in the predicted meteorological data, combined with the solar azimuth and the solar altitude angle, an improved dynamic programming algorithm is used to optimize and solve the optimal angle sequence of the photovoltaic tracking bracket within the future preset time period, including: Acquire motion parameters of the photovoltaic tracking bracket, the motion parameters including a minimum rotation angle, a maximum rotation angle, and a rotation power consumption coefficient; establish an initial angle state space based on the minimum rotation angle and the maximum rotation angle; Calculating an air quality index based on the solar azimuth angle and the solar altitude angle, and calculating direct radiation intensity based on the air quality index and solar radiation intensity; and multiplying the direct radiation intensity by the rotational power consumption coefficient to obtain a theoretical optimal power generation power; Establishing a state density function based on the theoretical optimal power generation; adaptively dividing the initial angular state space according to the state density function, subdividing the region where the state density is higher than a preset density threshold, and obtaining an optimized angular state space; calculating the maximum power generation corresponding to each angular state in the optimized angular state space; Calculating a temperature correction coefficient based on the temperature data and the temperature coefficient in the predicted meteorological data, multiplying the temperature correction coefficient by the maximum power generation to obtain the actual power generation; and constructing a power generation cost function by weighted combination of the actual power generation and the maximum power generation; In the optimized angle state space, the power generation cost function is substituted into the dynamic programming recursive equation, and the optimal state value at each moment is obtained through iterative calculation; and the optimal angle sequence within a future preset time period is generated based on the optimal state value.

5. The method according to claim 4, characterized in that In the optimized angular state space, the power generation cost function is substituted into a dynamic programming recursive equation, and the optimal state value at each moment is obtained through iterative calculation; Generating an optimal angle sequence within a future preset time period based on the optimal state value includes: Calculate the normalized difference between the actual generated power and the predicted generated power to obtain a power prediction error; substitute the power prediction error into an exponential decay function and multiply it by a basic discount rate to obtain a time-varying discount factor; add the negative value of the generated power cost function to the maximum value of the time-varying discount factor and the state value at the next moment to construct a dynamic programming recursive equation; Calculating a current angle state value based on the dynamic programming recursive equation, performing weighted averaging of the current angle state value and the historical angle state value according to a preset learning rate to obtain an updated angle state value; calculating a relative error between the updated angle state value and the historical angle state value, and when the relative error is less than a preset convergence threshold, recording the updated angle state value as the final state value; The angle that maximizes the final state value is selected from the optimized angle state space as the optimal angle at that moment; whether the optimal angle difference between adjacent predicted moments exceeds a preset maximum angle change rate is determined; if so, the optimal angle is corrected based on the maximum angle change rate; the corrected optimal angles are connected in sequence to generate an optimal angle sequence for the photovoltaic tracking bracket within a future preset time period.

6. The method according to claim 1, characterized in that Generating an angle adjustment instruction for the photovoltaic tracking bracket according to the optimal angle sequence includes: The motion constraint parameters of the photovoltaic tracking bracket are obtained, wherein the motion constraint parameters include a maximum adjustment speed, a maximum acceleration, and a minimum adjustment time interval; the optimal angle sequence is subjected to trajectory smoothing based on the motion constraint parameters to generate a transition angle sequence that satisfies the maximum adjustment speed and the maximum acceleration; the transition angle sequence is subjected to time discretization sampling according to the minimum adjustment time interval, and the transition angle sequence after the discretization sampling is converted into an angle adjustment instruction.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images