High-altitude photovoltaic module intelligent monitoring optimization system based on internet of things

Abstract

The application discloses a high-altitude photovoltaic module intelligent monitoring optimization system based on an Internet of Things, and belongs to the technical field of photovoltaic systems. The system comprises the following steps: obtaining disturbance factors by arranging multiple types of sensors, constructing a disturbance vector and fusing it into a comprehensive disturbance score as a unified environmental input of the whole system; combining the actual output power of the module, evaluating the theoretical output capacity in a lightweight model deployed on an edge node, and calculating the operation health score based on the deviation; further combining the score trend and disturbance intensity to judge the classification index; generating a structured response strategy according to the state label, disturbance type and score fluctuation, and specifically defining the action type, execution time and priority; completing the action execution by the edge node, and collecting the power change for feedback scoring. The scheme solves the problems of insufficient monitoring granularity, inaccurate state recognition, extensive control response and strong communication dependence in the plateau power station through module-level disturbance modeling, non-static state judgment and multi-factor decision generation.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic system technology, and in particular relates to an intelligent monitoring and optimization system for high-altitude photovoltaic modules based on the Internet of Things. Background Technology

[0002] With the large-scale deployment of photovoltaic power stations in high-altitude regions such as Tibet, ensuring the long-term stable operation of photovoltaic modules in extreme environments has become a key challenge in practical engineering. High-altitude regions present several natural conditions significantly different from those in plains areas, such as high annual ultraviolet radiation intensity, large diurnal temperature variations, frequent snow accumulation that is difficult to clear naturally, thin air, complex terrain, and weak communication infrastructure. These factors collectively lead to abnormal phenomena in photovoltaic modules during operation, such as power fluctuations, performance degradation, and short-term shading, which traditional centralized monitoring systems struggle to detect in a timely manner. Current monitoring methods primarily rely on inverter or combiner box data analysis, typically at the string level, failing to accurately assess the status of individual modules. Furthermore, monitoring models are often based on experience from low-altitude areas, lacking the ability to model typical disturbance factors such as ultraviolet radiation, wind erosion, and snow accumulation, resulting in inaccurate status assessments. On the other hand, existing systems mostly employ preset thresholds and fixed rules for control responses, unable to adaptively adjust control strategies based on the source of disturbance, their own operating status, and environmental trends. This often leads to slow response, control errors, or resource waste when facing disturbances at high altitudes. Meanwhile, communication instability or interruption is common in remote power plants. Existing solutions are unable to independently complete status perception, judgment and action execution under offline conditions, resulting in delayed fault handling and loss of system energy efficiency.

[0003] To address these issues, we propose an IoT-based intelligent monitoring and optimization system for high-altitude photovoltaic modules. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of slow response, control errors or resource waste in the prior art, and to propose an intelligent monitoring and optimization system for high-altitude photovoltaic modules based on the Internet of Things.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] The IoT-based intelligent monitoring and optimization system for high-altitude photovoltaic modules includes:

[0007] The data acquisition module is configured to acquire raw data; standardize the raw data to obtain a perturbation factor vector; and generate a perturbation score by weighted fusion of the perturbation factor vector.

[0008] The state assessment module is configured to collect the actual power of the component; construct a state scoring model; input the disturbance score, the actual power of the component, and the disturbance factor vector into the state scoring model, output a health score, and perform a moving average processing on the health score to obtain a smooth score.

[0009] An anomaly detection module is configured to construct a scoring normalization function based on the health score, the smoothing score, and the perturbation score; output a classification index value, and construct and output a status label based on the classification index value according to a preset rule.

[0010] The strategy generation module is configured to construct a strategy generation function based on the state label, the disturbance factor vector, and the health score. The strategy generation function constructs a disturbance response mapping matrix for the plateau photovoltaic environment, which is used to map the component state and the current environmental disturbance conditions to specific control response levels. It outputs a set of structured response actions, which includes control strategy type, strategy priority, execution timing, local energy consumption required for the response action, and action execution time or duration.

[0011] The execution recording module is configured to parse and execute the structured response action set, and to quantitatively record the operational changes of each component before and after the execution of the response action. The quantitative recording is completed by collecting the power change amplitude of the component within a fixed time window and recording it as a feedback score.

[0012] Preferably, the raw data includes ultraviolet irradiance, component surface temperature, snow thickness, and wind speed.

[0013] Preferably, the standardization process in the data acquisition module is to subtract the historical minimum irradiance value of the region from the current acquired value, and then divide by the difference between the historical maximum and minimum values ​​of the region.

[0014] Preferably, the state scoring model is a single-layer nonlinear network, which is trained using historical operating data before deployment and runs inference with static weights after deployment.

[0015] Preferably, the preset rules in the anomaly detection module are as follows:

[0016] When the classification index value is greater than the first threshold and the difference between the smoothed score and the health score is less than 0.01, the system is operating normally and no intervention is required.

[0017] When the classification index value is between the first threshold and the second threshold, and the difference between the smooth score and the health score is less than 0.01, it indicates a slight environmental impact.

[0018] When the classification index value is less than the second threshold and the snow perturbation term in the perturbation factor vector is less than the preset threshold, the output is a decrease in component performance.

[0019] When the classification index value is less than the second threshold and the snow disturbance term in the disturbance factor vector is greater than the preset threshold, a serious fault or occlusion is detected, requiring an immediate response.

[0020] Preferably, the strategy generation function also incorporates a score fluctuation term, which describes the dynamic changes in the component health score over the past k periods.

[0021] Preferably, the row index of the disturbance response mapping matrix corresponds to the state label of each component; the column index is composed of disturbance factor vectors encoded according to a preset segmentation rule.

[0022] Preferably, the value of each cell in the disturbance response mapping matrix represents the level of the system's recommended response strategy. This value corresponds one-to-one with the strength, priority, and degree of intervention of the control action, as detailed below:

[0023] A value of 0 indicates no response; only the status is recorded.

[0024] 1 indicates a minor parameter adjustment;

[0025] 2 indicates that the abnormal state has been recorded but no action will be triggered yet;

[0026] 3 indicates that the system has initiated a light cleaning or electrical adjustment operation;

[0027] 4 indicates that the component should be bypassed immediately or included in the manual maintenance plan.

[0028] In summary, the technical effects and advantages of this invention are as follows: This solution acquires disturbance factors such as ultraviolet radiation, temperature, wind speed, and snow accumulation by deploying multiple types of sensors, constructs disturbance vectors, and fuses them into a comprehensive disturbance score, which serves as a unified environmental input for the entire system. Then, combined with the actual output power of the components, its theoretical output capability is evaluated in a lightweight model deployed at the edge nodes, and an operational health score is calculated based on the deviation. Furthermore, the scoring trend and disturbance intensity are combined to determine classification indicators, thereby converting the current state of each component into a discrete label to indicate whether it is a short-term environmental disturbance, continuous degradation, or a serious fault. Subsequently, the system generates a structured response strategy based on the state label, disturbance type, and score fluctuation, specifically defining the action type, execution timing, and priority, and introducing a response penalty term based on the disturbance background to reduce the probability of erroneous actions. Finally, the edge nodes complete the action execution and collect power changes for feedback scoring, forming a complete control closed loop that can operate offline. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the system structure in this invention. Detailed Implementation

[0030] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0031] like Figure 1 As shown, the IoT-based intelligent monitoring and optimization system for high-altitude photovoltaic modules includes: a data acquisition module, which is configured to acquire raw data; standardize the raw data to obtain a disturbance factor vector; and generate a disturbance score by weighted fusion of the disturbance factor vector.

[0032] The state assessment module is configured to collect the actual power of the component; construct a state scoring model; input the disturbance score, the actual power of the component, and the disturbance factor vector into the state scoring model, output a health score, and perform a moving average processing on the health score to obtain a smooth score.

[0033] An anomaly detection module is configured to construct a scoring normalization function based on the health score, the smoothing score, and the perturbation score; output a classification index value, and construct and output a status label based on the classification index value according to a preset rule.

[0034] The strategy generation module is configured to construct a strategy generation function based on the state label, the disturbance factor vector, and the health score. The strategy generation function constructs a disturbance response mapping matrix for the plateau photovoltaic environment, which is used to map the component state and the current environmental disturbance conditions to specific control response levels. It outputs a set of structured response actions, which includes control strategy type, strategy priority, execution timing, local energy consumption required for the response action, and action execution time or duration.

[0035] The execution recording module is configured to parse and execute the structured response action set, and to quantitatively record the operational changes of each component before and after the execution of the response action. The quantitative recording is completed by collecting the power change amplitude of the component within a fixed time window and recording it as a feedback score.

[0036] The complete implementation process of this system is as follows:

[0037] Step 1, Data Acquisition Module:

[0038] In high-altitude regions, the operation of photovoltaic modules is affected by a variety of environmental factors, including ultraviolet radiation, temperature difference, snow depth, and wind speed. Because these environmental factors exhibit strong variability and regional differences in high-altitude areas, it is essential to monitor these disturbance factors in real time through a precise sensing system and convert them into standardized data to provide accurate input for subsequent module health assessments and optimized control.

[0039] The goal of this step is to acquire disturbance factor data in the plateau environment through sensing devices, and to standardize and model the data to provide high-quality input data for subsequent component status assessment and response optimization.

[0040] First, sensors deployed on the photovoltaic modules will monitor environmental factors in real time and collect the following data:

[0041] Ultraviolet irradiance intensity U(t): This data is acquired by an ultraviolet radiation sensor, typically the UV-1 type ultraviolet radiometer. This sensor is used to measure ultraviolet irradiance intensity in real time, with units of W / m². 2 Furthermore, it provides data support for subsequent analysis based on real-time irradiance changes.

[0042] Component surface temperature T(t): The temperature of the component surface is measured by a K-type thermocouple sensor, and the unit is °C. This sensor has high-precision temperature measurement capability and is suitable for high-altitude, low-pressure, and large-temperature-difference environments.

[0043] Snow thickness H(t): Snow depth data is acquired using a LiDAR-3000 sensor. This sensor calculates snow thickness in meters by reflecting laser light, accurately reflecting the actual thickness of snow on the component surface.

[0044] Wind speed V(t): Wind speed data is measured using an ultrasonic anemometer (WS-300 type), with units of m / s. This sensor has strong resistance to wind and sand interference and is suitable for wind speed measurement in high-altitude areas.

[0045] All the raw data collected will be transmitted to the system's edge computing nodes for real-time processing via wireless communication modules (such as the LoRaWAN protocol).

[0046] After data collection, these data need to be standardized because different environmental factors (such as ultraviolet radiation and wind speed) have different dimensions and numerical ranges. We use the min-max normalization method to ensure that perturbation factors with different units and dimensions can be fused and calculated in the same model.

[0047] In actual operation, the original ultraviolet irradiance U(t) is acquired in real time by ultraviolet sensors deployed on each component, and this value is typically expressed in W / m².2 Taking a certain plateau station as an example, its historical ultraviolet irradiance variation range is approximately 200 W / m². 2 Up to 1500W / m 2 To facilitate unified modeling with other environmental factors, we standardized this raw value. Specifically, we subtracted the historical minimum irradiance value U of the region from the current collected value U(t). min Then divide by the difference between the historical maximum and minimum values ​​of the region (U max -U min The result obtained is the ultraviolet radiation normalization factor E. U (t). This value is a dimensionless floating-point number, theoretically ranging from 0 to 1, representing the relative position of the current irradiance level within the historical range.

[0048] Using the same method, we also standardized the other three disturbance factors: normalized the current temperature T(t), snow thickness H(t), and wind speed V(t) using their corresponding historical extreme value ranges, and finally obtained E respectively. T (t), E H (t) and E V (t), these three and E U Together, these (t) constitute the perturbation factor vector E(t). This vector is used uniformly as the input to subsequent models and serves as a standard description of perturbations in the plateau environment, ensuring that different physical quantities can be fused and analyzed within the same logical system without causing modeling bias due to differences in units or orders of magnitude.

[0049] Next, the standardized perturbation factor vector E(t) is used as input to generate a perturbation score E for assessing the overall impact of the current environment. model (t). This score aims to unify the influence of multiple perturbation factors such as ultraviolet radiation, temperature, snow depth, and wind speed into a continuous value through weighted fusion, for use in subsequent models. In constructing E... model At time (t), the system assigns a weight coefficient to each perturbation factor in the vector. For example, ultraviolet radiation corresponds to weight w1, temperature to w2, snow cover to w3, and wind speed to w4. The standardized value of each factor is multiplied by its corresponding weight and then summed to obtain the perturbation score E at that time point. model(t). This score itself is a floating-point number between 0 and 1, reflecting the strength of the overall environmental disturbance. The weights are set according to the different factors affecting the component's operation. For example, in strong ultraviolet regions, such as areas above 3500 meters in altitude on the Qinghai-Tibet Plateau, ultraviolet radiation significantly increases the impact on component performance aging and power fluctuations, so w1 is usually set to a higher value. Temperature varies greatly daily in plateau regions but has a relatively small impact on short-term operation, so the value of w2 can be appropriately reduced. Weights can also be based on historical operating data through empirical regression analysis and allow for dynamic fine-tuning according to site environmental characteristics during system deployment. Through the weighted fusion design of this disturbance scoring model, data of different physical dimensions are unified into comparable inputs, which not only improves the model's training efficiency and prediction accuracy but also gives subsequent health scoring and response decisions a stronger environmental awareness, making it particularly suitable for applications with highly time-varying component operating states and complex disturbance patterns in high-altitude environments.

[0050] In this step, the selection of weighting coefficients directly affects the calculation of the environmental disturbance score. Therefore, fine-tuning the weighting coefficients using field monitoring data and historical feedback information can improve the model's adaptability and accuracy. Due to differences in ultraviolet radiation, temperature differences, and snow cover in different plateau regions, the adjustment of weighting coefficients will be more regionally specific. Finally, the standardized environmental disturbance score E... model (t) will provide key input data for the generation of state assessment models and control strategies in subsequent steps.

[0051] Step 2, Status Assessment Module:

[0052] In step 1, we conducted a comprehensive disturbance sensing of the high-altitude environment where the photovoltaic modules are located, and output the disturbance factor vector E(t) = [E U (t),E T (t),E H (t),E V [(t)] and its weighted fusion result E model (t) is used to describe the overall impact of the current environment on the component's operation. However, this disturbance information itself cannot directly determine whether the component is currently "operating well" or "degrading". The core task of this step is to compare these environmental awareness data with the actual power behavior of the components, thereby constructing a stable and deployable operational status assessment model and quantifying the operational health level of each component, providing accurate input for subsequent anomaly identification and optimized control.

[0053] We introduce a component health scoring function CHS. i (t) is used to measure the operating state of photovoltaic module i at time t. Input data includes: perturbation score E. model(t), the normalized perturbation vector E(t), and the actual power output P of the component. i (t) (in W), the latter is collected by a combination of current and voltage sensors at the back end of each module, with a sampling period of 60 seconds, and transmitted to the edge node for processing through the measurement module at the module end. To avoid using a uniform static reference power baseline for the entire module field (which is unsuitable due to large differences in terrain and orientation), we use a disturbance-aware power estimation model to construct the "expected power" of each module under the environmental conditions, and define a scoring function accordingly.

[0054] Considering the limitations of edge computing capabilities in model deployment, we chose a single-layer nonlinear network to implement this prediction model. The input is a five-dimensional vector [E(t), E model The output is a floating-point value representing the predicted power. The model is trained on historical data before system deployment and runs inference with static weights after deployment, without online learning. To ensure the model's robustness against disturbances, we introduce a disturbance sensitivity factor λ(t) to structurally correct the scores under specific extreme conditions.

[0055] Component Status Rating CHS i The complete calculation of (t) is as follows:

[0056]

[0057] Where ε is a minimal number to prevent division by zero (e.g., 10). -6 ), where λ(t) is a perturbation-sensitive adjustment term used to enhance the model's response to environmental changes in scenarios such as high reflectivity and extreme low temperatures. This adjustment term is defined as follows:

[0058] λ(t)=1+α·(E H (t) 2 +β·E U (t))

[0059] Where E H (t) is the snow accumulation factor, E U (t) represents the ultraviolet irradiance factor, and α and β are empirical adjustment coefficients used to enhance the penalty for power deviation under conditions of "high snow cover" or "strong ultraviolet radiation". For example, in the early morning in winter in Tibet, when ultraviolet irradiance increases but snow cover is still thick, traditional scoring models are prone to misjudging as "insufficient normal output", while this structure can automatically relax the scoring conditions and avoid false alarms.

[0060] Rating results CHS i The value of (t) ranges between [0,1], where a value closer to 1 indicates a healthier component state, and a value closer to 0 indicates a significant deviation from the theoretical output. To enhance the system's stability under frequent fluctuations in high-altitude climates, we...i (t) is processed by moving average to obtain a smooth score. A sliding window length of 10 minutes is recommended to ensure that the scoring curve is more robust in trend identification.

[0061] This step ultimately outputs two results: a single-moment score (CHS). i (t) and sliding score sequence This will be used for abnormal state classification in step 3. Compared with traditional methods that judge component abnormalities based on fixed thresholds, this scoring structure introduces a disturbance environment perception term and a dynamic correction term λ(t). Under special high-altitude environments (such as strong ultraviolet radiation, snow reflection, and temperature difference shocks), it can more realistically restore the component's operating state, improve the system's diagnostic accuracy, and reduce the false alarm rate. At the same time, it has strong edge deployment feasibility and is a core functional module that is truly compatible with the structure of this patented system.

[0062] Step 3, Anomaly Detection Module:

[0063] In step 2, we base our analysis on the environmental disturbance vector E(t) and the disturbance score E. model (t) A component state evaluation model was constructed, which outputs the health score CHS of each component at the current moment. i (t), and its moving average Used to judge the output performance of this component in the short term and trend dimensions. Although CHS i (t) and While component states can be quantified, they remain floating-point indices and cannot be directly used as discrete conditions for triggering control strategies. Therefore, they need to be further transformed into discrete state labels (CCLs) with semantic representation capabilities. i (t). This step aims to transform continuous state scores into category labels with clear control significance. Considering the characteristics of high-altitude areas, such as volatile disturbances and high risk of misjudgment, an adaptive classification criterion is designed to improve the system's environmental adaptability and execution accuracy.

[0064] In traditional power plants, component anomalies are often detected using fixed thresholds, such as CHS. i (t) values ​​below a certain threshold are directly classified as "abnormal". However, this method is highly susceptible to failure in high-altitude environments. Taking Tibet as an example, strong ultraviolet radiation causes the module's operating temperature to rise rapidly during the day, leading to voltage fluctuations. Simultaneously, snow cover or frost often occurs in the mornings, causing a short-term drop in module output. Such environmental disturbances are characterized by seasonality, non-uniformity, and regionality; using a fixed scoring threshold would significantly increase the false alarm rate.

[0065] To address this issue, we introduce a perturbation-sensitive dynamic scoring normalization function θ. i (t), this function is based on CHS i(t), and environmental disturbance score E model (t) is constructed, and the formula is as follows:

[0066]

[0067] Where γ and η are adjustment factors, with empirical values ​​set in the range of [0.3, 0.6]. ΔCHS i (t) represents CHS i (t) and its moving average The difference between the scores and the threshold values ​​reflects the stability of the ratings. When the scores fluctuate significantly, the system automatically reduces its sensitivity in classification to prevent misclassifying short-term fluctuations as anomalies; conversely, when the scores remain consistently low with minimal fluctuations, the system increases its recognition accuracy and identifies potential problems more quickly. This difference is achieved by using ΔCHS... i (t) is incorporated into the calculation process of the classification index, θ i (t) can not only reflect the trend level of the score itself, but also has the dynamic response capability to the intensity of disturbance and the stability of the score, thus constructing a classification mechanism that is more adapted to the extreme climate conditions of the plateau.

[0068] In constructing θ i After (t), we construct the state label CCL based on its value. i (t), outputting four component running states:

[0069] CCL i (t) = 0: The system is functioning normally and requires no intervention;

[0070] CCL i (t) = 1: Slight environmental impact (e.g., light frost / light snow accumulation);

[0071] CCL i (t) = 2: Component performance deteriorates (e.g., aging, microcracks, etc.);

[0072] CCL i (t) = 3: Critical fault or obstruction, requiring immediate response.

[0073] The classification logic uses a combinational judgment logic: when θ i (t)≥0.85 and ΔCHS i (t) 2 When θ < 0.01, it is considered normal; when θ i If (t)∈[0.6,0.85] and the score fluctuation is less than the set threshold, it is judged as a slight environmental impact; if θ i (t) < 0.6 and fluctuates greatly, and E H If (t)(snow accumulation disturbance term) is less than 0.2, it is considered a performance degradation; if E H(t) is greater than 0.2 and θ i If (t) is low at the same time, it is judged as severe snow accumulation or malfunction.

[0074] For example, at a high-altitude photovoltaic site after a snowstorm, the system detected the current health score (CHS) of a certain module. i (t) is 0.58, the average sliding score over the past 10 minutes. The value is 0.7, and the environmental disturbance score E*model(t) at that moment is 0.68. Combined with the score difference ΔCHS i (t) (i.e., the difference between the two is 0.12) and the set adjustment coefficient, the system substitutes these values ​​into the classification index θ i In the calculation logic of (t), the result obtained is approximately 0.52, significantly lower than the threshold required for normal judgment. Upon further examination of the perturbation factor, the snow accumulation factor E at that moment was found to be... H If (t) is greater than 0.2, it indicates that the component is very likely still in a partially snow-covered state. Based on the system classification rules, the component state is ultimately marked as CCL. i If (t) = 3, which indicates a serious anomaly, the system will automatically include it in the high-priority response queue and prepare to execute the corresponding control strategy.

[0075] Ultimately, CCL i (t) is the output of this step, a well-defined and easily integrated state label that drives the generation of the control response in the next step. Its generation process organically combines scoring trends, fluctuation levels, and disturbance conditions, overcoming the static defects of fixed threshold classification methods. It is a classification strategy structure specifically designed for the high-altitude environment monitoring scenario of this patented system.

[0076] Step 4, Strategy Generation Module:

[0077] In step 3, we have generated discretized state labels CCL based on the component scores and disturbance conditions. i (t), where i represents the component number, t represents the current time, and the label value is {0, 1, 2, 3}, representing "normal", "minor disturbance", "performance degradation", and "serious failure" respectively. This label only has a descriptive function in the system and cannot directly drive the execution of optimization control. Therefore, the goal of this step is to base the optimization control on CCL. i (t), combining the perturbation factor vector E(t) from step 1 and the component health score CHS constructed in step 2. i (t), design a response strategy generation mechanism D with scene awareness and adaptive adjustment capabilities. i (t) is used to guide component-level optimization operations.

[0078] The unique characteristics of the plateau environment dictate that the response strategy in photovoltaic arrays cannot adopt the traditional "static rule-fixed operation" mode. Taking snow disturbance as an example, the output drops sharply when the module is covered by snow, but no immediate manual intervention is required. The system can recover simply by adjusting the tracking points or marking the cleaning priority. As for the slight attenuation under wind and sand cover, if it occurs in the remote offline area, it can be handled by grouping and reducing weights, local bypassing, or strategic waiting. Therefore, the response strategy must have two capabilities: (1) based on the status label CCL i (t) Make an action judgment; (2) Make "action intensity fine-tuning" based on the perturbation combination E(t) and the characteristics of the scoring curve to form a flexible adjustment mechanism and avoid inefficient rigid strategies.

[0079] To this end, we designed a disturbance-aware response policy generation function, which has the following form:

[0080] D i (t)=ρ(CCL i (t),Φ i (t),E(t))

[0081] Where D i (t) is the set of structured response actions; ρ(·) is the policy function; Φ i (t) is the score fluctuation term, used to describe the component health score CHS. i The dynamic change of (t) over the past k periods can be expressed by the formula:

[0082]

[0083] This rating fluctuation term reflects the "stability" of the component rating within a short time window. If Φ i A large (t) indicates that the component is unstable and it is not advisable to immediately trigger drastic control actions. A "wait-and-see" strategy should be adopted. If the fluctuation item is small and the score is low, it indicates that the component is stable but poor and clear intervention measures should be adopted.

[0084] Inside the policy function ρ(·), we construct a perturbation response mapping matrix oriented towards the plateau photovoltaic environment. This matrix is ​​used to map component states to current environmental disturbance conditions and then to specific control response levels. The row indices of this matrix correspond to the state label (CCL) of each component. i (t), the column index is determined by the perturbation factor E. H (t), E U (t) and E V(t) is coded according to preset segmentation rules, representing typical disturbance combinations at the current moment. The value of each cell in the matrix represents the recommended response strategy level of the system. This level value corresponds one-to-one with the strength, priority, and degree of intervention of the control action. For example, a value of 0 indicates no response and only the status is recorded; 1 indicates a slight parameter adjustment; 2 indicates that the abnormal status is recorded but no action is triggered temporarily; 3 indicates that the system initiates a light cleanup or electrical adjustment operation; and 4 indicates that the component is immediately bypassed or included in the manual maintenance plan. This matrix was designed in the early stage of system deployment based on field engineering experience, historical operating data, and environmental disturbance simulation results. During system operation, an online fine-tuning mechanism is retained to adapt to the operating characteristics and disturbance patterns of different sites, thereby achieving adaptive upgrades at the strategy level.

[0085] In the response strategy structure, each Dx(t) contains the following fields:

[0086] Control strategy type (e.g., MPPT weight reduction, bus bypass, snow removal command);

[0087] Policy priority (mapped from CCL) i (t) and Φ i (combinations of t);

[0088] Execution timing (e.g., immediate execution, delayed execution, waiting for the recovery window);

[0089] Estimate the local energy consumption required for the response action (for energy consumption constraint modeling during scheduling);

[0090] Action execution time or duration (used to avoid frequent repetitive actions in high-reflection scenarios).

[0091] To further improve the stability of system policy execution and avoid frequent triggering of control actions when disturbances are not significant, we introduced an innovative control term called the disturbance background penalty regularization term Penalty during policy scheduling. i (t). This term is not directly used to control execution, but rather participates in adjusting the policy strength during policy generation, particularly playing a role in "weakening the response amplitude" under non-critical environmental disturbances. Specifically, this penalty term consists of two parts: one part comes from the snow accumulation factor E in the environmental disturbance. H The squared term of (t) is used to measure the degree of passive shading of the component under extreme snow cover conditions; another part comes from the scoring fluctuation term Φ. i (t), which has been defined above as the component score CHS. i (t) and moving average The short-term fluctuation intensity reflects the stability of the component's state. The system uses two adjustment coefficients, λ1 and λ2, to weight and fuse these two factors. λ1 controls the weight of environmental factors, and λ2 controls the impact of score fluctuations. The default values ​​of both are less than 1. If E H (t) is larger or Φ i A significant increase in (t) indicates that the current component is in a complex or unstable state, and the system will significantly improve Penalty. i The value of (t) determines whether the current control strategy is "high energy consumption and low benefit," thus proactively delaying, downgrading, or canceling the action. This mechanism is particularly important in high-altitude areas, effectively preventing the system from repeatedly initiating control commands under critical conditions such as morning temperature fluctuations and incomplete snow melting, thereby extending component lifespan, reducing energy consumption, and minimizing the consumption of limited communication and human maintenance resources.

[0092] The final output D i (t) will be directly interpreted and executed by the edge device, no longer relying on instructions from the central server. The results can be used by downstream modules such as component MPPT adjustment, electrical bypass signal issuance, component snow removal task queue sorting, and cleaning plan generation. Compared to traditional fixed-logic strategy systems, this solution establishes a four-dimensional fusion path between classification labels, disturbance factors, score fluctuations, and strategy actions, improving the intelligence and scenario adaptability of strategy generation, and is particularly suitable for the "unmanned operation and maintenance of high-altitude photovoltaic systems" target scenario in this patent.

[0093] Step 5, Execution Record Module:

[0094] In step 4, we generated a structured response strategy D for each photovoltaic module. i (t), whose content includes fields such as control type, priority, execution delay, and action duration. The goal of this step is to actually parse and execute these policy instructions in the edge device, and to quantify and record the operational changes of each component before and after the execution of the response action, forming a feedback value F. i (t) serves as an important basis for evaluating system behavior. This step marks the transition of the system from "judgment" to "execution," ultimately forming a complete closed loop of "diagnosis-response-feedback."

[0095] Each edge control unit (such as a low-power embedded master controller with a Cortex-M4 core) is connected to the acquisition and control loop of the controlled component via a local Modbus or RS485 bus. For each D... i (t), the edge unit first parses its key fields:

[0096] Control types, such as MPPT parameter adjustment, bus bypass, vibration snow removal signal, control delay, etc.

[0097] The delay time for instruction execution;

[0098] Action duration;

[0099] Information such as whether interruption is allowed and whether postponement is permitted.

[0100] Taking the response strategy of component i as an example, if If executed immediately, the edge device will control the inverter MPPT parameter adjustment module through the PWM modulation signal to reduce the maximum power point tracking voltage setting in the combiner box of this component.

[0101] In component response strategy D i (t) After being parsed and executed by the edge device, the system needs to evaluate the actual effect of the action. To determine whether the operation is effective, we set a fixed sampling window length Δt, typically 180 seconds, that is, continuously monitoring the output power change of the component within 3 minutes after the control command is executed. The power value P of the component before executing the strategy. i (t) is acquired by the voltage and current sampling modules connected to the edge node; after the strategy execution is completed, the actual power output of the same component is measured again and recorded as P. i (t+Δt). Subsequently, the system calculates the power change caused by the operation based on the power difference between the two time points and the reference power value at the current time, and defines the result as the component's feedback score F. i (t).

[0102] Feedback rating F i (t) reflects the relative increase or decrease in the component's output capability due to the current action. If P i (t+Δt) is significantly higher than P i (t) indicates that the action was effective; for example, clearing snow or optimizing MPPT settings led to power generation recovery. In this case, F i (t) is positive; a larger value indicates a more significant increase. Conversely, if the power decreases or remains almost unchanged, then F... i A negative or near-zero (t) indicates poor performance. To avoid calculation anomalies at extremely low power levels (e.g., a denominator of zero), the system introduces a very small positive number ε (e.g., 10) in the evaluation. -6 This serves as a safety protection item for the power benchmark. This feedback score not only evaluates the effectiveness of the current strategy but also provides a basis for subsequent optimization of strategy weights and adjustment of action triggering logic, making it one of the key output variables for the system to achieve "closed-loop self-learning capability."

[0103] For example, after the component performs a "snow removal vibration operation", the power drops from P. i (t) = 160W increases to P i (t+Δt)=208W, then This represents a 30% increase. If the power does not increase significantly or even decreases, the feedback is negative, indicating poor performance or that the disturbance has not been resolved.

[0104] Feedback F i (t) will interact with the original control strategy D i (t), Component's current tag CCL i (t) are packaged together and stored in a local log queue, and uploaded to the central platform via MQTT or CoAP protocol when communication conditions permit. This data can be used not only for later manual operation and maintenance analysis, but also as a data source for subsequent model retraining.

[0105] The most crucial aspect of this step lies in defining the power change rate as a feedback indicator and independently collecting and evaluating data based on a standard time window. This constructs a response execution and evaluation mechanism that can operate independently at the edge, detached from the central system. This mechanism not only adapts to high-altitude, remote, unmanned scenarios but also possesses engineering advantages such as lightweight structure, flexible deployment, and quantifiable traceability, significantly enhancing the system's closed-loop control capabilities and effectiveness.

[0106] The technical solutions in the above-described embodiments of this application have at least the following technical effects or advantages: This solution acquires disturbance factors such as ultraviolet radiation, temperature, wind speed, and snow accumulation by deploying multiple types of sensors, constructs disturbance vectors, and fuses them into a comprehensive disturbance score, which serves as a unified environmental input for the entire system; then, combined with the actual output power of the components, its theoretical output capability is evaluated in a lightweight model deployed at the edge nodes, and a health score is calculated based on the deviation; further, the classification index is determined by combining the score trend and disturbance intensity, thereby converting the current state of each component into a discrete label to indicate whether it is a short-term environmental disturbance, continuous degradation, or a serious fault; subsequently, the system generates a structured response strategy based on the state label, disturbance type, and score fluctuation, specifically defining the action type, execution timing, and priority, and introducing a response penalty term based on the disturbance background to reduce the probability of erroneous actions; finally, the edge nodes complete the action execution and collect power changes for feedback scoring, forming a complete control closed loop that can operate offline.

[0107] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A high-altitude photovoltaic module intelligent monitoring optimization system based on Internet of Things, characterized in that, include: A data acquisition module is configured to acquire raw data and perform standardization processing on the raw data to obtain a perturbation factor vector. The disturbance factor vector is weighted and fused to generate a disturbance score; the original data includes ultraviolet irradiance, component surface temperature, snow thickness, and wind speed; the standardization process is to subtract the historical minimum irradiance value of the region from the current collected value, and then divide by the difference between the historical maximum and minimum values ​​of the region. A status assessment module, which is configured to collect the actual power of the component; The state score model is constructed; the disturbance score, the actual power of the component, and the disturbance factor vector are input into the state score model, and a health score is output; the health score is processed by a sliding average to obtain a smooth score; a disturbance sensitive factor is introduced into the state score model , which is used for structural correction of the score under extreme environment, and the specific implementation is as follows: ； wherein is a snow factor, is an ultraviolet radiation factor, and are empirical adjustment factors to enhance the degree of penalty for power deviation under high snow or strong ultraviolet conditions; An anomaly detection module is configured to construct a scoring normalization function based on the health score, the smoothness score, and the perturbation score; output a classification index value; and construct and output a status label based on the classification index value according to preset rules; the preset rules are as follows: When the classification index value is greater than the first threshold and the square of the difference between the smooth score and the health score is less than 0.01, the system is operating normally and no intervention is required. When the classification index value is between the first threshold and the second threshold, and the square of the difference between the smooth score and the health score is less than 0.01, it indicates a slight environmental impact. When the classification index value is less than the second threshold and the snow accumulation factor in the disturbance factor vector is less than the preset threshold, the output is a decrease in component performance. When the classification index value is less than the second threshold and the snow accumulation factor in the disturbance factor vector is greater than the preset threshold, a serious fault or occlusion is detected, requiring an immediate response. The strategy generation module is configured to construct a strategy generation function based on the state label, the disturbance factor vector, and the health score. The strategy generation function constructs a disturbance response mapping matrix for the plateau photovoltaic environment, used to map the component state to the current environmental disturbance conditions as specific control response levels. The strategy generation function also introduces a score fluctuation term, which describes the dynamic changes of the health score over the past k periods. The row index of the disturbance response mapping matrix corresponds to the state label of each component; the column index is encoded by combining the disturbance factor vector according to a preset segmentation rule. The strategy generation function also introduces a disturbance background penalty regularization term, which includes the square of the snow accumulation factor from the environmental disturbance and a score fluctuation term. The score fluctuation term is defined as the short-term fluctuation intensity of the health score and the smoothing score, reflecting the stability of the component state. When the current component is in a complex or unstable state, increase the value of the penalty regularization term to make the current control strategy judged as high energy consumption and low benefit, thereby actively delaying, downgrading or canceling the action; output a structured response action set, which includes control strategy type, strategy priority, execution timing, local energy consumption required for the response action, action execution time or duration. The execution recording module is configured to parse and execute the structured response action set, and to quantitatively record the operational changes of each component before and after the execution of the response action. The quantitative recording is completed by collecting the power change amplitude of the component within a fixed time window and recording it as a feedback score.

2. The IoT-based intelligent monitoring and optimization system for high-altitude photovoltaic modules according to claim 1, characterized in that, The state scoring model is a single-layer nonlinear network. It is trained using historical operating data before deployment and runs inference with static weights after deployment.

3. The IoT-based intelligent monitoring and optimization system for high-altitude photovoltaic modules according to claim 1, characterized in that, The value of each cell in the disturbance response mapping matrix represents the level of the system's recommended response strategy. This value corresponds one-to-one with the strength, priority, and degree of intervention of the control action, as detailed below: A value of 0 indicates no response; only the status is recorded. 1 indicates that parameter adjustments are being made; 2 indicates that the abnormal state has been recorded but no action will be triggered yet; 3 indicates that the system has initiated a light cleaning or electrical adjustment operation; 4 indicates that the component should be bypassed immediately or included in the manual maintenance plan.

Images