Wind-sand erosion prediction and adaptability evaluation method for photovoltaic panel in Sagomean region

Abstract

The invention discloses a sandstorm erosion prediction and adaptability evaluation method for a photovoltaic panel in a Saggob area. The method comprises the following steps: collecting sandstorm environment parameters and panel state parameters by arranging monitoring equipment such as a micro meteorological station and an unmanned aerial vehicle aerial photography system to form a multi-source data basis; key erosion dynamics characteristics are extracted based on the wind sand dynamics principle; performing an accelerated test by using a wind-sand erosion simulation device to quantify the relationship between the material wear rate and the performance attenuation; constructing a time sequence deep learning network to establish an erosion evolution prediction model; and finally, according to a prediction result, evaluating an adaptability level and formulating a targeted protection strategy. According to the method, multi-source monitoring data and a machine learning technology are creatively fused, high-precision prediction and dynamic evaluation of the wind-sand erosion process of the photovoltaic panel are realized, prediction accuracy and mechanism interpretability are remarkably improved, a scientific basis can be provided for active operation and maintenance of a power station, and the method is suitable for popularization and application. And the long-term stability and the operation and maintenance efficiency of the photovoltaic power station in a severe environment are effectively enhanced.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic panel technology, and in particular to a method for predicting and assessing the wind and sand erosion of photovoltaic panels in desert and Gobi areas. Background Technology

[0002] Inner Mongolia and other arid, sandy regions are core areas for the construction of large-scale national photovoltaic bases. However, the harsh wind and sand environment in these regions has led to severe wind and sand erosion problems for photovoltaic panels. Wind and sand erosion not only directly wears down the surface materials of the panels, but also affects the heat dissipation and power generation efficiency of the panels through the accumulation of sand and dust.

[0003] Chinese invention patent application number 202411288648.3 discloses a smart photovoltaic power station intelligent inspection and management system and method. The method includes: acquiring supporting soil image information, photovoltaic panel supporting image information, and terrain slope; based on this, determining the risk of photovoltaic panel misalignment; acquiring local photovoltaic heat island information, analyzing the supporting soil image information and local photovoltaic heat island information to determine supporting soil change information; acquiring wind and sand climate information, analyzing the wind and sand climate information and supporting soil change information to determine erosion change information; based on the photovoltaic panel misalignment risk, and according to the supporting soil change information and erosion change information, predicting the angular misalignment development of the photovoltaic panels to determine the photovoltaic panel misalignment trend information; and visually outputting the photovoltaic panel misalignment trend information for maintenance personnel reference. This enables photovoltaic panels potentially affected by negative environmental impacts to be maintained in a timely manner, reducing the photovoltaic panel failure rate.

[0004] In the desert and Gobi regions, existing technologies face significant bottlenecks: First, their models struggle to quantify the multi-physics micro-erosion and wear mechanisms of windblown sand on panels and supporting structures; second, predictions rely primarily on macroscopic environmental parameters, failing to fully consider the region's unique sand grain creep, migration, and suspension patterns, as well as the interference effects of photovoltaic array layout on near-surface windblown sand, resulting in insufficient prediction accuracy; finally, current assessments focus on the static determination of past migration or erosion, lacking dynamic predictions of erosion accumulation processes and material performance degradation, thus failing to provide forward-looking adaptive assessments and protection guidance. Therefore, a shift from monitoring macroscopic phenomena to quantifying and dynamically predicting the evolution of microscopic erosion mechanisms is needed to improve the accuracy and foresight of early warning systems. Summary of the Invention

[0005] This application provides a method for predicting and assessing the adaptability of photovoltaic panels to wind and sand erosion in desert and Gobi areas. It addresses the shortcomings of traditional methods in terms of prediction accuracy, mechanism characterization, and assessment system, and provides a scientific basis for the planning, design, and operation and maintenance strategies of photovoltaic power plants.

[0006] This application provides a method for predicting and assessing the wind and sand erosion of photovoltaic panels in desert areas, including: S1 collects wind and sand environmental parameters and panel status parameters to form a multi-source data foundation by deploying micro weather stations, dust collectors, drone aerial photography systems and panel surface sensors in photovoltaic power stations; S2, based on the principles of wind and sand dynamics, extracts key erosion dynamic features from multi-source data to quantify the erosion mechanism; S3 utilizes a wind and sand erosion simulation device to conduct accelerated erosion tests on different types of photovoltaic panel materials, quantifying the relationship between material wear rate and performance degradation under different wind and sand conditions; S4, based on multi-source data, erosion mechanism and the relationship between wear rate and performance degradation, constructs a training dataset, and uses a time series deep learning network to build an evolutionary model to predict the wear depth of the panel and the thickness of the dust cover in the future dynamic time window; S5, based on the prediction results, combined with panel performance thresholds and operation and maintenance goals, assesses the adaptability level and then formulates targeted protection.

[0007] Preferably, the extraction of key erosion dynamics features includes: calculating the impact kinetic energy of sand particles impacting a unit panel area per unit time using aeolian physics formulas based on monitored wind speed data and sand particle size distribution; deriving the number of sand particles impacting the panel per unit time to determine the impact frequency by using wind speed, dust concentration, and sand particle size distribution and the relationship between sand flux and the impacted area of ​​the panel; calculating the dust deposition rate per unit area based on a deposition model by combining dust concentration, local wind speed corrected for photovoltaic array interference, and panel tilt angle; and deploying auxiliary anemometers at key locations of the photovoltaic array to measure actual wind speed changes, while using computational fluid dynamics to simulate the flow field structure under typical wind conditions based on the three-dimensional model of the power station, in order to calculate the rate of change of wind speed at specific points to quantify the degree of change in the local flow field.

[0008] Preferably, the quantification of the relationship between material wear rate and performance degradation under different wind and sand conditions includes: using a wind and sand erosion simulation test bench to conduct accelerated erosion tests on photovoltaic panel samples within a pre-set range of wind and sand conditions; controlling key parameters such as wind and sand flow velocity, sand content, and impact angle during the test, and recording the erosion duration and cumulative impact energy; measuring the change in light transmittance of the samples and calculating the loss rate using a spectrophotometer, observing the surface micromorphology and quantifying the wear depth using a scanning electron microscope, and measuring the output power attenuation rate using an IV curve tester; establishing empirical relationship models between light transmittance loss rate and cumulative impact kinetic energy, and between output power attenuation rate and wear depth based on the test data; and determining the characteristic coefficients in each empirical relationship model through statistical analysis to complete the quantitative characterization of the material performance degradation law.

[0009] Preferably, the construction of the evolutionary model for predicting the wear depth and dust cover thickness of the panel within a future dynamic time window specifically includes: aligning and fusing aeolian environmental parameters, dynamic characteristics, and laboratory damage data to construct a training dataset; using a time-series deep learning network as the core model architecture to process the training dataset to capture temporal dependence and spatial variability; training the model using historical data and optimizing network parameters to minimize prediction error; and outputting the prediction results of the wear depth and dust cover thickness of the panel surface within the future dynamic time window, wherein the model performance is evaluated using an error index.

[0010] Preferably, the targeted protection measures formulated after assessing the adaptability level specifically include: comprehensively evaluating the panel erosion state based on the output of the evolution prediction model using multiple indicators, determining the adaptability level by comparing the predicted values ​​with preset performance thresholds, and matching corresponding optimized protection measures from a preset protection strategy knowledge base according to the adaptability level and the predicted problem characteristics.

[0011] Preferably, step S3, which quantifies the relationship between material wear rate and performance degradation under different wind and sand conditions, further includes: S31, simultaneously collects topographic elevation data, seasonal meteorological sequences and microenvironmental wind and sand flow field data, and establishes a spatiotemporal fusion database; S32, develop an algorithm to correlate topographic parameters with seasonal indicators, extract the modulation characteristics of topographic factors on wind and sand activities under different seasonal backgrounds, including the variation law of topographic effects during the alternation of dry and rainy seasons; S33, Establish a photovoltaic array erosion prediction model that considers both topographic and seasonal factors, and optimize the model's adaptability during seasonal transitions through cross-validation. S34 integrates short-term weather forecasts and long-term seasonal trend data, runs a coupled model to output the spatiotemporal distribution of panel erosion risk, and provides key early warnings for high-risk areas with overlapping seasonal transition periods and complex terrain. S35 generates zoned panel protection schemes based on terrain features and seasonal characteristics.

[0012] Preferably, the establishment of the spatiotemporal fusion database includes: acquiring elevation point cloud data of the photovoltaic power station area using airborne lidar mapping technology; generating a digital terrain model after denoising and gridding; calculating key terrain parameters such as slope and aspect based on the model; deploying a network of micro-meteorological stations to monitor meteorological parameters; acquiring seasonal meteorological sequences through long-term observation; processing outliers and imputing missing values ​​in the collected data to maintain the integrity of the time series; deploying three-dimensional anemometers and dust sensors at key terrain points to monitor the characteristics of wind and sand flow fields, and achieving time synchronization with terrain data and meteorological data; integrating and fusing terrain elevation data, meteorological sequences, and wind and sand flow field data through spatiotemporal indexing to establish a spatiotemporal fusion database that supports multi-source data query and analysis.

[0013] Preferably, the extraction of the modulation characteristics of topographic factors on wind and sand activities under different seasonal backgrounds specifically involves: extracting topographic parameters and seasonal indicators from a spatiotemporal fusion database and performing normalization preprocessing on the data; establishing a quantitative relationship between topographic parameters and seasonal indicators through correlation analysis and regression models, including using linear correlation analysis to assess the degree of correlation between topographic factors and seasonal wind and sand activities; extracting the modulation characteristics of topography on wind and sand activities based on the analysis results, including topographic acceleration coefficients and seasonal modulation factors; and applying time series analysis methods to track the dynamic changes of topographic effects during seasonal transitions and identify key turning points.

[0014] Preferably, the optimization of the model's adaptability during seasonal transitions further includes: constructing a time-series prediction model that integrates terrain features and seasonal indicators, using a long short-term memory network as its basic architecture, and introducing a feature weighting mechanism to dynamically adjust the contribution of different features; extracting training data from a spatiotemporal fusion database and related features, and performing feature fusion and time-series alignment processing; training the model using historical data, and adjusting network parameters by optimizing the loss function; evaluating model performance using a time-series cross-validation method, and dynamically adjusting hyperparameters based on the validation results to enhance the model's predictive adaptability during seasonal transitions.

[0015] Preferably, the high-risk area specifically includes: collecting short-term weather forecast data and long-term seasonal trend data, and achieving multi-source data fusion with topographic data through spatiotemporal alignment and interpolation processing; running a topographic-seasonal dual-factor prediction model, calculating the erosion risk score of each grid unit based on the fused data, and generating a time-series risk distribution map through spatial mapping; and extracting high-risk areas exceeding a preset threshold based on the risk distribution results.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: By integrating wind and sand dynamics mechanisms, multi-source monitoring data, and machine learning models, high-precision prediction and dynamic assessment of wind and sand erosion processes of photovoltaic panels in desert and Gobi areas have been achieved. The technical solution significantly improves prediction accuracy and mechanism interpretability, enabling proactive early warning of erosion evolution and providing sufficient time window for proactive operation and maintenance of power plants. At the same time, the scientific and comprehensive adaptability assessment system comprehensively considers panel material characteristics, environmental stress, and economic factors, outputting quantitative levels and targeted protection strategies to directly guide power plant design and operation and maintenance decisions, effectively enhancing the long-term stability and operation and maintenance efficiency of photovoltaic power plants in harsh environments.

[0017] By integrating seasonal dynamic characteristics with complex terrain parameters, an adaptive prediction model was developed, significantly improving the accuracy and reliability of wind and sand erosion prediction for photovoltaic panels in desert and Gobi areas. This technical solution effectively addresses abrupt wind and sand activity during the transition between dry and rainy seasons and the local flow field effects under complex terrain, enabling accurate early warning for high-risk areas. Advantages include reduced prediction errors, shorter operation and maintenance response times, and enhanced long-term stability of power plants in harsh environments through zoned protection strategies, directly supporting the economical and efficient operation and maintenance of photovoltaic panels. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for predicting and assessing the adaptability of photovoltaic panels to wind and sand erosion in desert areas, according to an embodiment of the present invention. Detailed Implementation

[0019] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Example 1: Figure 1 This is a flowchart illustrating a method for predicting and assessing the adaptability of photovoltaic panels to wind and sand erosion in desert areas, according to an embodiment of the present invention.

[0022] like Figure 1 As shown, a method for predicting and assessing the wind and sand erosion of photovoltaic panels in desert and Gobi areas includes the following steps: S1 collects wind and sand environmental parameters and panel status parameters to form a multi-source data foundation by deploying micro weather stations, dust collectors, drone aerial photography systems, and panel surface sensors in photovoltaic power stations.

[0023] Specifically, miniature weather stations are deployed inside the photovoltaic power station and in open areas upwind, equipped with high-precision anemometers (accuracy ±0.1 m / s) and dust concentration sensors (range 0-2000 μg / m³). 3 The system includes gradient dust collection cups (for collecting dust samples at different heights). A drone aerial photography system (equipped with a multispectral camera and infrared thermal imager) and panel surface sensors (such as laser scanners) are deployed simultaneously to ensure coverage of key locations such as the first row of the array.

[0024] Continuous monitoring of wind speed, wind direction, temperature, humidity, and dust concentration was conducted, with data acquisition frequency once per minute to ensure real-time accuracy. Samples from gradient dust collection cups were periodically collected, and a laser particle size analyzer was used to determine the particle size distribution and obtain statistical characteristics of particle size.

[0025] Drones conduct aerial photography weekly: multispectral images are used to invert the thickness of dust cover on the panel surface through spectral reflectance characteristics; infrared thermal imagers detect back panel temperature anomalies (such as hot spots). A handheld laser scanner is used monthly to perform close-range scans of typical panels, quantifying surface micro-roughness (such as arithmetic mean roughness Ra) to characterize the degree of wear.

[0026] The collected raw data is cleaned (outliers or noise are removed), aligned (time series synchronization) and integrated, and stored in a unified data management platform to provide a structured dataset for subsequent analysis.

[0027] S2 extracts key erosion dynamic features from multi-source data based on the principles of wind and sand dynamics to quantify the erosion mechanism.

[0028] Among them, the erosion dynamics characteristics include sand grain impact kinetic energy, impact frequency, sand and dust deposition rate, and the degree of change in the local flow field of the photovoltaic array.

[0029] Specifically, based on monitored wind speed data and sand particle size distribution (from laser particle size analysis of S1), the total kinetic energy of sand particles impacting a unit panel area per unit time is calculated using the Bagnold wind physics formula. First, the mass of the sand particles is calculated using the particle size distribution, and then the kinetic energy value is calculated in conjunction with the wind speed data. The formula for calculating the sand particle mass is as follows: , The kinetic energy of a single sand grain is represented by joules (m); the mass of the sand grain is represented by kilograms (kg), calculated using the grain size distribution and density (density is typically taken as approximately 2650 kg / m³ for quartz sand). 3 ); v represents wind speed (unit: meters per second), taken from monitoring data.

[0030] The number of sand particles impacting a panel per unit time is derived using wind speed, dust concentration, and sand particle size distribution. An empirical relationship between wind speed and sand flux is established by statistically analyzing the duration and high-frequency data of wind and sand activity. The calculation method involves combining the impact frequency with the sand flux formula; the empirical relationship is as follows: f represents the impact frequency per unit area; Q represents the sand flux, which is obtained through a regression model of wind speed and dust concentration; A represents the impact area of ​​the panel.

[0031] Combining dust concentration, local wind speed (corrected for photovoltaic array interference), and panel tilt angle, the dust accumulation rate per unit area is calculated based on a sedimentation model. Panel tilt angle affects sedimentation efficiency through geometric relationships. The formula is: , Dust deposition rate (unit: kg / m³) 2 •s); C represents dust concentration (unit: kg / m³) 3 ), taken from sensor data; Local wind speed (unit: m / s), corrected for array interference; Indicates the panel tilt angle (unit: degrees). This reflects the effect of dip angle on sedimentation.

[0032] Auxiliary anemometers were deployed at key locations on the photovoltaic array to measure actual wind speed changes. Simultaneously, based on a 3D model of the power plant, CFD software was used to simulate the flow field structure under typical wind conditions, calculating the rate of change of wind speed at specific points. The quantification method involved outputting the rate of change of wind speed through CFD simulation. Δv represents the degree of change in the local flow field (unit: %). Indicates the measured or simulated wind speed at key points of the array; This indicates the wind speed in an open area upwind.

[0033] The wind speed range used in the calculations is typically 5–25 m / s, based on the percentile definition of historical monitoring data (e.g., above the 90th percentile indicates high erosion risk). The dust concentration range is 0–2000 μg / m³, referencing the range of the reference sensor. 3 High concentration (>1000 μg / m³) 3 This corresponds to a high deposition rate. A change in the flow field exceeding ±20% is considered a significant disturbance; the definition method is based on the allowable error range between CFD simulation and actual measurements.

[0034] S3 utilizes a wind and sand erosion simulation device to conduct accelerated erosion tests on different types of photovoltaic panel materials, quantifying the relationship between material wear rate and performance degradation under different wind and sand conditions.

[0035] Specifically, a wind erosion simulation test bench was used to control the wind and sand flow velocity within the range of 10-20 m / s, the sand content within the range of 0.1%-1%, and the impact angle within the range of 15°-90°, in order to cover the typical wind and sand intensity in the Gobi Desert region. The samples included different types of photovoltaic panel materials (such as glass and backsheet coating) to ensure that the initial state of the sample surfaces was consistent.

[0036] The samples were subjected to periodic erosion under controlled conditions, and variables such as erosion duration and cumulative impact energy were recorded. The experimental design included multiple control groups, such as fixed wind speed with varying sand concentration, or fixed impact angle with varying flow velocity, to isolate the influence of parameters.

[0037] Before and after the experiment, and after different periods, three key measurements were performed on the samples: the change in transmittance was measured using a spectrophotometer to calculate the loss rate; the surface microstructure was observed using a scanning electron microscope (SEM) to quantify the wear depth; and the output power attenuation rate was measured using an IV curve tester.

[0038] By correlating experimental data with erosion conditions, an empirical damage model was constructed with cumulative impact kinetic energy and erosion time as independent variables, and light transmittance loss rate and power attenuation rate as dependent variables.

[0039] In quantifying the relationship between wear rate and performance degradation, an empirical model is fitted based on experimental data. For example, the relationship between light transmittance loss rate and cumulative impact kinetic energy can be expressed as: ΔT represents the transmittance loss rate (unit: %), which is defined as the relative decrease in transmittance before and after the test; The transmittance attenuation coefficient (unit: % / J) related to the material is derived from experimental data through regression analysis. The cumulative impact kinetic energy (unit: joule) is calculated as follows: ,in This represents the kinetic energy of a single impact (based on the formula for the S2 step), and Δt is the erosion time interval. Another key model is the relationship between output power attenuation and wear depth, which can be expressed as: , This indicates the output power attenuation rate (unit: %), which is the percentage of power decrease measured by the IV curve tester. The power attenuation coefficient (unit: % / μm) depends on the material type and corrosion conditions and is obtained through experimental fitting. The wear depth (unit: μm) is obtained through surface morphology analysis using SEM images, such as using the arithmetic mean roughness (Ra) as a proxy. The above model is semi-empirical and needs to be validated with laboratory data. The coefficients in the formula (such as...) and It needs to be determined from multiple sets of experiments using statistical methods such as the least squares method.

[0040] S4, based on multi-source data, erosion mechanisms, and the relationship between wear rate and performance degradation, constructs a training dataset and uses a time-series deep learning network to build an evolutionary model that predicts the wear depth of panels and the thickness of dust cover in future dynamic time windows.

[0041] Specifically, the aeolian environmental parameters (such as wind speed and dust concentration time series) collected in step S1, the dynamic characteristics extracted in step S2 (such as sand particle impact kinetic energy and deposition rate), and the laboratory damage data obtained in step S3 (such as the relationship between material wear rate and performance degradation) are aligned and fused. Preprocessing includes data cleaning, normalization, and sequence segmentation, such as using historical data sequences from the past 30 days as a basis to ensure temporal consistency.

[0042] A time-series deep learning network is used as the core, with priority given to Long Short-Term Memory (LSTM) networks or their variants (such as ConvLSTM) to effectively capture temporal dependence and spatial variability. The input features in the model design include historical environmental sequences, dynamic characteristics, and material properties. The prediction targets are defined as the surface wear depth and dust cover thickness of the panel over the next 7, 30, and 90 days.

[0043] In describing LSTM networks, the core update equation is used to process sequential data. The mathematical formula is as follows: in, , , These represent the activation values ​​of the input gate, forget gate, and output gate, respectively. denoted by sigmoid, and tanh represents the hyperbolic tangent function; , , , Represents the weight matrix; , , , Represents the bias vector; Indicates the hidden state at the previous moment; This represents the input characteristics at the current moment (such as wind speed or impact kinetic energy). Indicates cell state, This indicates the hidden state at the current moment.

[0044] A portion (e.g., 70%) of historical data is used as the training set, with the remainder used for validation and testing. During training, the mean squared error (MSE) or mean absolute percentage error (MAPE) is used as the loss function, and the network parameters are optimized using the backpropagation algorithm. To enhance physical interpretability, the laboratory damage model from step S3 is used as a constraint, such as by adding a regularization term to the loss function, to ensure that the prediction results conform to the material wear behavior.

[0045] The Mean Absolute Percentage Error (MAPE) is used as a key metric when evaluating model performance. The mathematical formula is as follows: MAPE stands for Mean Absolute Percentage Error (unit: %). Indicates the number of predicted samples; Indicates the actual observed value; This represents the model's predicted values. The model output is a predicted panel condition for a specific future time window, including wear depth and dust cover thickness.

[0046] S5, based on the prediction results, combined with panel performance thresholds and operation and maintenance goals, assesses the adaptability level and then formulates targeted protection.

[0047] Specifically, the predicted panel conditions for specific future time periods (such as 7 days, 30 days, and 90 days) are obtained from the evolutionary prediction model trained in step S4. These predictions include surface wear depth, dust cover thickness, and the resulting rate of decline in power generation efficiency. These predictions need to be validated for consistency.

[0048] Based on the prediction results, combined with preset panel performance thresholds and power plant operation and maintenance goals, a comprehensive evaluation of multiple indicators is conducted. The evaluation criteria integrate predicted wear depth, light transmittance loss, power generation efficiency reduction, and impact on structural safety (such as wind and sand loads). The evaluation process involves comparing each predicted value with its corresponding threshold and mapping it to a predefined suitability level (A / D level). For example, if all predicted indicators are below the acceptable threshold, the evaluation is level A; if any indicator is close to or exceeds the critical value, it is classified as level B, C, or D according to its severity.

[0049] During the evaluation process, the fitness level mapping is based on multi-indicator threshold comparisons. Mathematically, this can be achieved through conditional logic, such as setting threshold ranges for each indicator. The fitness level (L) can be determined as follows: Where L represents the fitness level (A, B, C or D); Indicates the predicted indicators (such as wear depth, light transmittance loss rate); , , This indicates the upper limit of the threshold for the corresponding level, based on panel material specifications and operational cost presets.

[0050] Based on the assessed adaptability level and the predicted main problems (such as whether dust accumulation or wear is the primary issue), the optimal strategy is matched from a pre-built knowledge base of protection strategies.

[0051] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By integrating wind and sand dynamics mechanisms, multi-source monitoring data, and machine learning models, high-precision prediction and dynamic assessment of wind and sand erosion processes of photovoltaic panels in desert and Gobi areas have been achieved. The technical solution significantly improves prediction accuracy and mechanism interpretability, enabling proactive early warning of erosion evolution and providing sufficient time window for proactive operation and maintenance of power plants. At the same time, the scientific and comprehensive adaptability assessment system comprehensively considers panel material characteristics, environmental stress, and economic factors, outputting quantitative levels and targeted protection strategies to directly guide power plant design and operation and maintenance decisions, effectively enhancing the long-term stability and operation and maintenance efficiency of photovoltaic power plants in harsh environments.

[0052] Example 2: In Example 1, when the photovoltaic panel wind erosion prediction model was applied to a desert region, evolutionary prediction was performed based on static data acquisition and a unified dynamic feature extraction method. However, this model failed to fully consider the coupled effects of seasonal abrupt changes (such as the suddenness of wind and sand activity during the transition between dry and rainy seasons) and complex terrain (such as the local flow field effects in mountains and hills), resulting in significant prediction biases during seasonal transitions or in areas with undulating terrain. Since seasonal climate change alters sand particle size distribution and wind erosion intensity, and topographic undulation modulates the movement patterns of near-surface wind erosion, a unified prediction model struggles to capture these dynamic heterogeneities, inevitably leading to insufficient generalization ability and local prediction inaccuracies. To improve the model's adaptability and accuracy in harsh environments, it is necessary to integrate seasonal indicators and topographic parameters simultaneously, optimizing and improving the prediction framework, thus leading to Example 2.

[0053] In some embodiments, to quantify the relationship between material wear rate and performance degradation under different wind and sand conditions, step S3 further includes: S31 synchronously collects topographic elevation data, seasonal meteorological sequences, and microenvironmental wind and sand flow field data to establish a spatiotemporal fusion database.

[0054] Specifically, airborne lidar mapping technology was used to scan the photovoltaic power station area at a resolution of 5 meters to acquire elevation point cloud data. Data processing included point cloud denoising and meshing to generate a digital terrain model (DTM), and calculation of key terrain parameters such as slope and aspect. Slope calculation was based on the rate of elevation change, estimated using the elevation difference of neighboring pixels; aspect was calculated using azimuth angles to reflect the terrain orientation. This process ensured high accuracy and spatial consistency of the terrain data.

[0055] A network of miniature weather stations was deployed to monitor parameters such as wind speed, wind direction, temperature, and humidity, with data acquisition frequency at once per minute. Seasonal sequences were obtained through long-term observations (e.g., three months each for the dry and rainy seasons) and stored as time-series datasets. Data preprocessing included outlier removal and missing value imputation; for example, for data points with discontinuous timestamps, linear interpolation was used to ensure sequence integrity. The linear interpolation formula was applied in time alignment. Where V(t) represents the interpolated value at time t, and Adjacent time points and The observed values. This allows for smooth alignment of meteorological data over time.

[0056] Three-dimensional ultrasonic anemometers and dust sensors are deployed at key topographic points (such as slope tops and valley bottoms) to monitor the velocity distribution and particle concentration of windblown sand, with a sampling frequency of 10Hz. Data is transmitted in real-time to a central platform for preliminary quality control, such as filtering high-frequency noise. Synchronous acquisition ensures that the timestamps of windblown sand data are consistent with those of topographic and meteorological data, providing a foundation for data fusion.

[0057] Topographic elevation data, meteorological sequences, and aeolian flow field data are integrated into a spatiotemporal database and organized using a time-space index. The fusion process includes data alignment (ensuring time synchronization and spatial coordinate consistency across all datasets) and normalization (eliminating dimensional differences). The database structure supports rapid querying and analysis, such as retrieving comprehensive data for specific time periods and regions using SQL or a spatiotemporal query language.

[0058] S32, develop an algorithm to correlate topographic parameters with seasonal indicators, and extract the modulation characteristics of topographic factors on wind and sand activities under different seasonal backgrounds, including the variation law of topographic effects during the alternation of dry and rainy seasons.

[0059] Specifically, topographic parameters (such as slope and aspect) and seasonal indicators (such as dry and rainy season classification and seasonal average wind speed) are extracted from the S31 spatiotemporal fusion database. Seasonal indicators are defined based on historical meteorological data; for example, the annual average wind speed sequence is divided into dry seasons (periods of higher wind speed) and rainy seasons (periods of lower wind speed) by month, and statistical characteristics (such as mean and variance) for each season are calculated. Data preprocessing includes normalization to eliminate dimensional differences and ensure consistency in subsequent correlation analyses.

[0060] Correlation analysis and regression models were designed to establish a quantitative relationship between topographic parameters and seasonal indicators. First, the Pearson correlation coefficient was used to assess the linear correlation between topographic factors and seasonal aeolian activity indicators (such as dust concentration). The Pearson correlation coefficient calculation formula is applied in the correlation analysis: Where r represents the correlation coefficient. and These represent topographic parameters (such as slope) and seasonal indicators (such as average wind speed during the dry season), respectively. and is the mean, and n is the number of samples. This coefficient quantifies the strength of the linear correlation between topography and season, and its value range is [-1, 1].

[0061] For nonlinear relationships, a multivariate regression model is introduced, for example, using terrain parameters as independent variables and seasonal wind and sand characteristics as dependent variables to construct a prediction equation.

[0062] Based on the results of the correlation algorithm, the modulation features of topography on aeolian activity are extracted, such as the topographic acceleration coefficient (the enhancing effect of slope on wind speed) and the seasonal modulation factor (the difference in topographic effects between dry and rainy seasons). Feature extraction is achieved by calculating residual analysis or effect sizes, for example, using the coefficients of a regression model to characterize the contribution of topographic factors. The modulation features are stored as feature vectors for subsequent pattern analysis.

[0063] Focusing on seasonal transition periods (such as the 30 days before and after the transition from dry to rainy season), time series analysis methods (such as sliding window correlation) are applied to track the dynamic changes in topographic effects. For example, the rolling correlation coefficient between topographic parameters and wind and sand activity during the transition from dry to rainy season is calculated to identify trends. The output of the pattern analysis includes topographic effect change curves and key inflection points, revealing the temporal evolution of modulation characteristics.

[0064] S33. Establish a photovoltaic array erosion prediction model that considers both topographic and seasonal factors, and optimize the model's adaptability during seasonal transitions through cross-validation.

[0065] Specifically, a time-series prediction model integrating topographic and seasonal features is constructed, based on the LSTM network in Example 1, with expanded input dimensions. Input features include topographic parameters (such as slope and aspect), seasonal indicators (such as seasonal classification codes), and historical aeolian environmental data (such as wind speed and dust concentration). The model output is the panel erosion state (such as wear depth) for a specific future time period (such as 7 days or 30 days). In the model design, a feature weighting mechanism is introduced, for example, dynamically adjusting the contribution of topographic and seasonal features through an attention layer to capture key changes during seasonal transitions.

[0066] Training data, including topographic parameter sequences, seasonal index time series, and erosion response data, was extracted from the spatiotemporal fusion database of S31 and the associated features of S32. Feature fusion employed a stitching method to align topographic and seasonal features with the time series, ensuring that each time step contained multidimensional information. Data was partitioned by season (e.g., dry season and rainy season datasets) to support season-specific analysis.

[0067] The model is trained using historical data, and parameters are optimized using mean squared error as the loss function. The mean squared error formula is applied during the training process: ,in, Represents the actual observed value. This represents the model's predicted value, where n is the number of samples. This function quantifies the prediction bias and minimizes the error through backpropagation. During training, a regularization term is added to prevent overfitting and ensure the model's generalization ability.

[0068] A time-series cross-validation method is employed to optimize the model's adaptability during seasonal transitions. Specifically, the data is divided into multiple folds in chronological order, for example, using the transition point between dry and rainy seasons as the dividing point, ensuring that each fold contains a complete seasonal cycle. Cross-validation calculates the average performance metric for each fold, such as the mean absolute percentage error (MAPE), through iterative training and validation to assess model stability. The MAPE formula is applied in validation: This indicator helps identify prediction biases in models during seasonal transitions and guides parameter adjustments.

[0069] Based on cross-validation results, model hyperparameters (such as learning rate and hidden layer dimension) are dynamically adjusted, with a focus on optimizing prediction performance during seasonal transitions. For example, during the alternation of dry and rainy seasons, the weight coefficients of seasonal features are increased, and the model is fine-tuned using the gradient descent algorithm. After optimization, the model is better able to capture the interaction between terrain and seasons, such as the weakening effect of wind speed acceleration on slopes during the dry season and its reduction during the rainy season.

[0070] S34 integrates short-term weather forecasts and long-term seasonal trend data, runs a coupled model to output the spatiotemporal distribution of erosion risk on the panel, and provides key early warnings for high-risk areas with overlapping seasonal transition periods and complex terrain.

[0071] Specifically, short-term weather forecast data (such as wind speed, wind direction, and precipitation probability forecasts for the next 7 days) and long-term seasonal trend data (such as statistical analysis of drought and rainy season patterns based on historical meteorological data) are collected. During data integration, a spatiotemporal alignment method is employed to ensure that the time series of weather forecasts are consistent with the spatial scale of seasonal trends. For example, interpolation techniques are used to downscale the forecast data to the grid cells of a photovoltaic array and overlay it with topographic data (such as slope and aspect). During the alignment process, for data with mismatched timestamps, linear interpolation is applied for synchronization to ensure that all data are consistent in the spatiotemporal dimensions.

[0072] Run the terrain-seasonal two-factor prediction model trained in S33, inputting the integrated data (including weather forecasts, seasonal trends, and terrain parameters). The model outputs a predicted future erosion state for each grid cell, such as abrasion depth and dust cover thickness. Based on the predictions, calculate an erosion risk score, for example, by quantifying the risk level using a weighted synthesis method. The risk scoring formula is applied in the assessment: Where R represents the risk score, D is the predicted wear depth (normalized value), S is the seasonal factor (e.g., 1.0 for the dry season and 0.7 for the rainy season, based on the intensity of historical wind and sand activity), and T is the terrain complexity factor (calculated based on slope). , , The weighting coefficients are determined through training with historical data. This formula integrates multiple factors into a single risk indicator, facilitating subsequent analysis.

[0073] Risk scores are mapped to a spatial grid to generate a time-series erosion risk distribution map. For example, the risk distribution for the next 7 days is output on a daily basis, and spatial interpolation methods (such as Kriging interpolation) are used to fill in unmonitored areas, forming a continuous surface. The distribution map highlights high-risk areas, such as areas overlapping with steep slopes during seasonal transitions (before and after the transition from dry to rainy seasons). Based on the risk distribution map, areas with scores exceeding a preset threshold are identified as key areas for early warning. The threshold is set with reference to historical erosion event data; for example, a percentile method is used to define a high-risk threshold. ,in, Indicates a high-risk threshold This represents the 90th percentile of the historical risk score. Early warning information includes the risk level, potential impact, and recommended actions, and is pushed to operations and maintenance personnel in real time via a decision support system.

[0074] S35 generates zoned panel protection schemes based on terrain features and seasonal characteristics.

[0075] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By integrating seasonal dynamic characteristics with complex terrain parameters, an adaptive prediction model was developed, significantly improving the accuracy and reliability of wind and sand erosion prediction for photovoltaic panels in desert and Gobi areas. This technical solution effectively addresses abrupt wind and sand activity during the transition between dry and rainy seasons and the local flow field effects under complex terrain, enabling accurate early warning for high-risk areas. Advantages include reduced prediction errors, shorter operation and maintenance response times, and enhanced long-term stability of power plants in harsh environments through zoned protection strategies, directly supporting the economical and efficient operation and maintenance of photovoltaic panels.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting and assessing the adaptability of photovoltaic panels to wind and sand erosion in desert and Gobi areas, characterized in that, include: S1 collects wind and sand environmental parameters and panel status parameters to form a multi-source data foundation by deploying micro weather stations, dust collectors, drone aerial photography systems and panel surface sensors in photovoltaic power stations; S2, based on the principles of wind and sand dynamics, extracts key erosion dynamic features from multi-source data to quantify the erosion mechanism; S3 utilizes a wind and sand erosion simulation device to conduct accelerated erosion tests on different types of photovoltaic panel materials, quantifying the relationship between material wear rate and performance degradation under different wind and sand conditions; S4, based on multi-source data, erosion mechanism and the relationship between wear rate and performance degradation, constructs a training dataset, and uses a time series deep learning network to build an evolutionary model to predict the wear depth of the panel and the thickness of the dust cover in the future dynamic time window; S5, based on the prediction results, combined with panel performance thresholds and operation and maintenance goals, assesses the adaptability level and then formulates targeted protection.

2. The method for predicting and assessing the wind erosion of photovoltaic panels in desert and Gobi areas as described in claim 1, characterized in that, The extraction of key erosion dynamics features includes: calculating the impact kinetic energy of sand particles impacting a unit panel area per unit time using aeolian physics formulas based on monitored wind speed data and sand particle size distribution; deriving the number of sand particles impacting the panel per unit time to determine the impact frequency by using wind speed, dust concentration, and sand particle size distribution and the relationship between sand flux and the impacted area of ​​the panel; calculating the dust deposition rate per unit area based on a deposition model by combining dust concentration, local wind speed corrected for photovoltaic array interference, and panel tilt angle; and deploying auxiliary anemometers at key locations of the photovoltaic array to measure actual wind speed changes, while using computational fluid dynamics to simulate the flow field structure under typical wind conditions based on the three-dimensional model of the power station to calculate the rate of change of wind speed at specific points to quantify the degree of change in the local flow field.

3. The method for predicting and assessing the wind erosion of photovoltaic panels in desert and Gobi areas as described in claim 1, characterized in that, The quantitative analysis of the relationship between material wear rate and performance degradation under different wind and sand conditions includes: using a wind and sand erosion simulation test bench to conduct accelerated erosion tests on photovoltaic panel samples within a pre-set range of wind and sand conditions; controlling key parameters such as wind and sand flow velocity, sand content, and impact angle during the test, and recording the erosion duration and cumulative impact energy; measuring the change in light transmittance of the samples using a spectrophotometer and calculating the loss rate; observing the surface micromorphology and quantifying the wear depth using a scanning electron microscope; measuring the output power attenuation rate using an IV curve tester; establishing empirical relationship models between light transmittance loss rate and cumulative impact kinetic energy, as well as empirical relationship models between output power attenuation rate and wear depth based on the experimental data; and determining the characteristic coefficients in each empirical relationship model through statistical analysis to complete the quantitative characterization of the material performance degradation law.

4. The method for predicting and assessing the wind erosion of photovoltaic panels in desert and Gobi areas as described in claim 1, characterized in that, The construction of the evolutionary model for predicting the wear depth and dust cover thickness of the panel within a future dynamic time window specifically includes: aligning and fusing aeolian environmental parameters, dynamic characteristics, and laboratory damage data to construct a training dataset; using a time-series deep learning network as the core model architecture to process the training dataset to capture temporal dependence and spatial variability; training the model using historical data and optimizing network parameters to minimize prediction error; and outputting the prediction results of the wear depth and dust cover thickness of the panel surface within the future dynamic time window, wherein the model performance is evaluated using an error metric.

5. The method for predicting and assessing the wind erosion of photovoltaic panels in desert and Gobi areas as described in claim 1, characterized in that, The targeted protection measures after assessing the adaptability level include: comprehensively evaluating the panel erosion state based on the output of the evolution prediction model, determining the adaptability level by comparing the predicted value with a preset performance threshold, and matching corresponding optimized protection measures from a preset protection strategy knowledge base according to the adaptability level and the predicted problem characteristics.

6. The method for predicting and assessing the wind erosion of photovoltaic panels in desert areas as described in claim 1, characterized in that, The step S3, which quantifies the relationship between material wear rate and performance degradation under different wind and sand conditions, further includes: S31, simultaneously collects topographic elevation data, seasonal meteorological sequences and microenvironmental wind and sand flow field data, and establishes a spatiotemporal fusion database; S32, develop an algorithm to correlate topographic parameters with seasonal indicators, extract the modulation characteristics of topographic factors on wind and sand activities under different seasonal backgrounds, including the variation law of topographic effects during the alternation of dry and rainy seasons; S33, Establish a photovoltaic array erosion prediction model that considers both topographic and seasonal factors, and optimize the model's adaptability during seasonal transitions through cross-validation. S34 integrates short-term weather forecasts and long-term seasonal trend data, runs a coupled model to output the spatiotemporal distribution of panel erosion risk, and provides key early warnings for high-risk areas with overlapping seasonal transition periods and complex terrain. S35 generates zoned panel protection schemes based on terrain features and seasonal characteristics.

7. The method for predicting and assessing the wind erosion of photovoltaic panels in desert and Gobi areas as described in claim 6, characterized in that, The establishment of the spatiotemporal fusion database includes: acquiring elevation point cloud data of the photovoltaic power station area using airborne lidar mapping technology; generating a digital terrain model after denoising and gridding; calculating key terrain parameters such as slope and aspect based on the model; deploying a network of micro-meteorological stations to monitor meteorological parameters; acquiring seasonal meteorological sequences through long-term observation; performing outlier processing and missing value imputation on the collected data to maintain the integrity of the time series; deploying three-dimensional anemometers and dust sensors at key terrain points to monitor the characteristics of wind and sand flow fields and achieving time synchronization with terrain and meteorological data; and integrating and fusing terrain elevation data, meteorological sequences, and wind and sand flow field data through spatiotemporal indexing to establish a spatiotemporal fusion database that supports multi-source data query and analysis.

8. The method for predicting and assessing the wind erosion of photovoltaic panels in desert and Gobi areas as described in claim 6, characterized in that, The extraction of the modulation characteristics of topographic factors on wind and sand activities under different seasonal backgrounds specifically involves: extracting topographic parameters and seasonal indicators from a spatiotemporal fusion database and performing normalization preprocessing on the data; establishing a quantitative relationship between topographic parameters and seasonal indicators through correlation analysis and regression models, including using linear correlation analysis to assess the degree of correlation between topographic factors and seasonal wind and sand activities; and extracting the modulation characteristics of topography on wind and sand activities based on the analysis results, including topographic acceleration coefficients and seasonal modulation factors. Time series analysis methods are used to track the dynamic changes in topographic effects during seasonal transitions and identify key turning points.

9. The method for predicting and assessing the wind erosion of photovoltaic panels in desert areas as described in claim 6, characterized in that, The optimization of the model's adaptability during seasonal transitions further includes: constructing a time-series prediction model that integrates terrain features and seasonal indicators, using a long short-term memory network as its basic architecture, and introducing a feature weighting mechanism to dynamically adjust the contribution of different features; extracting training data from a spatiotemporal fusion database and related features, and performing feature fusion and time-series alignment processing; training the model using historical data, and adjusting network parameters by optimizing the loss function; evaluating model performance using a time-series cross-validation method, and dynamically adjusting hyperparameters based on the validation results to enhance the model's predictive adaptability during seasonal transitions.

10. The method for predicting and assessing the wind erosion of photovoltaic panels in desert areas as described in claim 6, characterized in that, The high-risk areas specifically include: collecting short-term weather forecast data and long-term seasonal trend data, and achieving multi-source data fusion with topographic data through spatiotemporal alignment and interpolation processing; running a topographic-seasonal dual-factor prediction model, calculating the erosion risk score of each grid unit based on the fused data, and generating a time-series risk distribution map through spatial mapping; and extracting high-risk areas exceeding a preset threshold based on the risk distribution results.

Images