Solar photovoltaic module array data acquisition method based on wireless Internet of Things

Abstract

The invention discloses a solar photovoltaic module array data acquisition method based on a wireless internet of things, and particularly relates to the technical field of solar photovoltaic. Configuring a wireless acquisition node for each photovoltaic module, wherein the nodes dynamically adjust the sampling frequency and the data fidelity level according to the illumination intensity, the historical volatility and the state of adjacent nodes; collecting multi-source operation data, extracting key features through weighted fusion and trend analysis, and performing preliminary dimension reduction in combination with a collaborative factor to generate a local feature data packet; data packets are divided into data layers according to content types and emergency degrees, and layered heterogeneous transmission is realized through a low-power-consumption wireless network; the edge gateway receives the data stream, reconstructs missing data by using the cooperation coefficient, and performs operation state prediction and anomaly correction through the graph neural network; the system realizes adaptive optimization of data acquisition and intelligent operation and maintenance of the system, and has the advantages of high efficiency, low consumption and the like.

Description

Technical Field

[0001] This invention relates to the field of solar photovoltaic technology, and more specifically to a method for acquiring data from a solar photovoltaic module array based on the wireless Internet of Things. Background Technology

[0002] With the adjustment of the global energy structure and the continuous development of renewable energy technologies, solar photovoltaic power generation, as a green and clean energy form, occupies an increasingly important position in the energy system. Photovoltaic module arrays are a key component of photovoltaic power generation systems, and their operating status directly affects the overall power generation efficiency and system stability. Therefore, how to efficiently and accurately collect and transmit operating data from photovoltaic module arrays has become a key focus of current research and application.

[0003] Traditional photovoltaic module data acquisition methods rely heavily on wired communication networks, which are complex, costly to maintain, and lack flexibility. This difficulty is further increased, especially in areas with complex terrain or widely distributed modules. Furthermore, the system is frequently affected by environmental interference, leading to discontinuous or distorted data acquisition, impacting operation and maintenance decisions and energy efficiency assessments.

[0004] With the rapid development of IoT and wireless communication technologies, applying wireless IoT technology to photovoltaic module data acquisition has become a feasible and efficient solution. This technology enables real-time acquisition and remote transmission of photovoltaic module data, improving the system's intelligence and automation. However, a complete, efficient, reliable, and clearly structured data acquisition method is still lacking, capable of meeting the real-time data acquisition needs of large-scale photovoltaic arrays in complex environments. Summary of the Invention

[0005] The purpose of this invention is to provide a data acquisition method for solar photovoltaic module arrays based on wireless Internet of Things, so as to overcome the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for acquiring data from a solar photovoltaic module array based on a wireless Internet of Things, comprising: Each module in the photovoltaic module array is configured with a wireless acquisition node, and the wireless acquisition node includes a unique identifier ID of the module, location information, self-learning weight parameters, and dynamic coordination factor. Each wireless acquisition node dynamically adjusts its sampling frequency and data fidelity level based on the current light intensity, historical sampling fluctuation rate, and the perception status of neighboring nodes, and performs multi-source data acquisition, including voltage, current, component surface temperature, and ambient light intensity. The collected multi-source data is processed using a weighted average and trend detection algorithm to extract key operational features. At the same time, the collaborative factors exchanged with neighboring nodes are fused to perform preliminary dimensionality reduction, generating a compressed local feature data package. The local feature data packets are divided into multiple data layers according to their content type and urgency, and are transmitted hierarchically and heterogeneously through a low-power wireless communication network. High-priority data is uploaded in real time, while low-priority data is uploaded after a delay. The edge gateway receives heterogeneous data streams from various nodes, uses the coordination coefficient to reconstruct missing data, employs a lightweight graph neural network model to predict and correct anomalies in the component's operating status, and outputs structured array operating data. The structured array's operating data is analyzed to generate an acquisition strategy optimization model. The sampling granularity, feature extraction parameters, and communication frequency of each wireless acquisition node are dynamically adjusted through downlink control commands.

[0007] Preferably, the dynamic adjustment of the sampling frequency and data fidelity level includes: Measure the current light intensity and calculate the rate of change of light intensity within a short time window; Based on the local storage of the rolling sample window, the variance and peak frequency of historical sampled values ​​are statistically analyzed to generate and output the historical sampled volatility index. The system receives operational status summaries from neighboring nodes via short-range low-power wireless broadcasting, and then weights and fuses these summaries with historical volatility to obtain a comprehensive stability assessment value. Based on the rate of change of illumination, historical volatility, and comprehensive stability assessment value, the sampling frequency, analog-to-digital conversion sampling accuracy, data compression ratio, and transmission priority are dynamically adjusted according to preset rules.

[0008] Preferably, the dynamic adjustment of sampling frequency, analog-to-digital conversion sampling accuracy, data compression ratio, and transmission priority includes: When the rate of change of illumination is greater than the set threshold, the sampling frequency is automatically increased; otherwise, the sampling frequency is decreased. Based on the adjusted sampling frequency, the sampling accuracy is improved when the sampling frequency is increased; Based on the current analog-to-digital conversion accuracy, the nodes dynamically adjust the data compression parameters; Based on the importance and load of the compressed data packets, high, medium, and low transmission priorities are assigned to the data packets, with high-priority data being uploaded via the wireless link first.

[0009] Preferably, the step of extracting key operational features from the collected multi-source data using a weighted average and trend detection algorithm includes: The collected voltage, current, component surface temperature and ambient light intensity data are subjected to dimension unification and normalization processing. Based on the degree of influence of each parameter on the output power stability of the photovoltaic module, a weight coefficient is assigned to each parameter, and the comprehensive operating characteristic value of the node is calculated. Within a set sliding time window, trend analysis is performed on continuous comprehensive characteristic values. By comparing the changing direction of the current characteristic value with the average value of the previous window, the upward, downward, or stable trend states can be identified. Based on the trend detection results, key operational feature sets are extracted, including power change trends, temperature drift direction, and light response sensitivity.

[0010] Preferably, the local feature data packet is divided into multiple data layers according to content type and urgency, including: When generating local feature data packets, classification labels are set according to the data source attributes and content types, which include three types of labels: real-time monitoring data, periodic statistical data, and abnormal alarm data. Based on the power fluctuation magnitude, temperature drift rate, and trend change direction contained in the data packet, an urgency index is calculated. When the index exceeds a set threshold, the data is marked as high priority. Data packets are stored in high, medium, and low priority buffer queues according to their classification identifier and urgency level. The high priority queue adopts a real-time sending strategy, while the medium and low priority queues enter the delayed aggregation waiting area.

[0011] Preferably, the edge gateway receives heterogeneous data streams from various nodes and uses a coordination coefficient to reconstruct the association of missing data, including: Receive heterogeneous data streams containing node identifiers, coordination coefficients, and timestamps from multiple wireless acquisition nodes, and build an index table based on the node identifiers; When a missing data node is detected, the missing data segment is reconstructed by interpolation fitting based on the coordination coefficient between the node and its neighboring nodes, selecting a dataset that is time-synchronized and highly coordinated from the historical data of the neighboring nodes. The reconstructed multi-node data is input into a lightweight graph neural network model, and the edge weights between nodes are defined by the synergy coefficient. Compare the predicted value with the actual collected value. When the deviation exceeds the set threshold, perform anomaly correction and mark the source of correction.

[0012] Preferably, the analysis of the structured array's operational data and the generation of an optimized acquisition strategy model include: Perform multi-dimensional statistical analysis on the structured array operation data uploaded by the edge gateway to extract power fluctuation amplitude, node data integrity rate and communication load intensity; Based on the analysis indicators, an optimization model for the acquisition strategy is established. The model uses the power fluctuation amplitude as a sensitive parameter and node energy consumption and communication load as constraints to output the sampling priority and parameter adjustment suggestions for each node. A downlink control instruction set is generated based on the optimization model results, including sampling granularity adjustment values, feature extraction frequency coefficients, and communication cycle correction amounts.

[0013] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention significantly improves the data acquisition efficiency and system stability of photovoltaic module arrays in dynamic environments by introducing a node-level adaptive sampling mechanism, a collaborative factor-driven data fusion strategy, and an edge computing intelligent completion and correction model. Compared with traditional fixed sampling schemes, this invention can dynamically adjust the sampling frequency, feature extraction accuracy, and data transmission strategy according to changes in illumination, historical volatility, and the status of neighboring nodes. This enables the priority perception and timely uploading of key feature information, while effectively suppressing redundant data transmission and reducing overall system energy consumption.

[0014] 2. This invention, by introducing a collaborative reconstruction algorithm and a lightweight graph neural network model into the edge gateway, enables the system to intelligently predict and correct operational status when data is lost or abnormally fluctuates, enhancing the system's fault tolerance to uncertainties such as sensor failures and communication interruptions. Combined with cloud-based strategy optimization and a downlink closed-loop control mechanism, it achieves intelligent configuration and dynamic adjustment of node parameters, further improving the intelligence level and scalability of the entire photovoltaic array operation and maintenance system. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] For examples, please refer to Figure 1 As shown in this embodiment, a data acquisition method for a solar photovoltaic module array based on wireless Internet of Things includes: Each module in the photovoltaic module array is configured with a wireless acquisition node, and the wireless acquisition node includes a unique identifier ID of the module, location information, self-learning weight parameters, and dynamic coordination factor. Each wireless acquisition node dynamically adjusts its sampling frequency and data fidelity level based on the current light intensity, historical sampling fluctuation rate, and the perception status of neighboring nodes, and performs multi-source data acquisition, including voltage, current, component surface temperature, and ambient light intensity. The collected multi-source data is processed using a weighted average and trend detection algorithm to extract key operational features. At the same time, the collaborative factors exchanged with neighboring nodes are fused to perform preliminary dimensionality reduction, generating a compressed local feature data package. The local feature data packets are divided into multiple data layers according to their content type and urgency, and are transmitted hierarchically and heterogeneously through a low-power wireless communication network. High-priority data is uploaded in real time, while low-priority data is uploaded after a delay. The edge gateway receives heterogeneous data streams from each node, uses the coordination coefficient to reconstruct missing data, employs a lightweight graph neural network model to predict and correct anomalies in the component's operating status, and outputs structured array operating data. The structured array's operating data is analyzed to generate an acquisition strategy optimization model. The sampling granularity, feature extraction parameters, and communication frequency of each wireless acquisition node are dynamically adjusted through downlink control commands.

[0019] In the data acquisition method for solar photovoltaic module arrays based on wireless Internet of Things proposed in this invention, to achieve accurate acquisition, intelligent transmission, and collaborative management of photovoltaic module operating data, a corresponding wireless acquisition node must first be configured for each module in the photovoltaic module array. This wireless acquisition node is physically bound to the module and is used to sense and collect the module's operating parameters, while simultaneously enabling collaborative sensing and dynamic adjustment with other nodes. To ensure the personalized, adaptive, and traceable operation of the system, each wireless acquisition node is pre-configured with the following key information or parameters: Component Unique Identifier (ID): Each photovoltaic module has a unique identifier, typically a factory serial number or a unique code within the system. This ID is bound one-to-one with the wireless data acquisition node for subsequent data uploading, status monitoring, and fault location. The system can use this ID to accurately map the operating data of a specific module, avoiding data confusion.

[0020] Location Information: Wireless acquisition nodes record their spatial location information within the entire photovoltaic array. This information can be represented using relative coordinates (such as row and column numbers) or absolute coordinates (such as GPS coordinates). The location information is used for subsequent data aggregation, array-level status analysis, local fault location, and coordinated scheduling of neighboring nodes, forming the foundation for regionalized sensing and management.

[0021] Self-learning weight parameters: The wireless acquisition node has a set of dynamically updatable self-learning weight parameters preset to quantitatively evaluate the importance or stability of the current node in system operation based on dimensions such as historical data quality, data volatility, and energy consumption. These parameters can serve as a reference for adjusting the node's sampling frequency, controlling communication frequency, and determining data priority.

[0022] Dynamic Coordination Factor: The dynamic coordination factor is used to characterize the degree of coordination between the current node and its neighboring nodes during data perception and transmission. This factor reflects the correlation and coupling strength between nodes, such as whether power change trends are consistent and whether illumination changes are synchronized.

[0023] To achieve accurate data acquisition and efficient transmission of photovoltaic modules under complex lighting conditions, this invention introduces a dynamic control mechanism in each wireless acquisition node. This mechanism allows the sampling frequency, analog-to-digital conversion accuracy, data compression ratio, and data transmission priority to be adjusted in real time according to environmental changes and node status. Specifically, this includes the following sequential steps: The wireless data acquisition node periodically measures the current light intensity using its built-in light sensor, with an initial sampling period of 1 minute. Within a set short sliding time window (e.g., 5 minutes), the node records multiple consecutive light intensity values. The rate of change of light intensity is defined as the ratio of the difference between the current light intensity value and the initial light intensity value within that time window to the time interval. That is: Rate of change of light intensity = (Current light intensity value - Initial light intensity value) / Time window width (unit time). This rate of change of light intensity reflects the degree of fluctuation in ambient light intensity over a short period. A high rate of change indicates potential rapid occlusion or cloud drift, requiring an increased data sampling frequency to capture abrupt changes.

[0024] Each node maintains a local rolling sample window to store operational data collected over a period of time, including voltage, current, surface temperature, and ambient light intensity. The window size can be set to 60 data samples. Statistical characteristics are calculated from the historical data within this window, particularly: Variance: representing the range of data fluctuations and reflecting system stability; Peak Count: recording the number of local maxima or minima in the collected data per unit time, used to determine the frequency of abrupt changes.

[0025] Nodes periodically receive operational status summaries from neighboring nodes via short-range low-power wireless broadcasting (such as BLE or ZigBee). These summaries include, but are not limited to, the current sampling frequency of neighboring nodes, the previously generated rate of change in illumination, and the most recently acquired anomalous fluctuation tags (such as anomalous jumps). After receiving summaries from at least two neighboring nodes, each node performs a weighted fusion based on its relative distance to its own location. For example, nodes that are closer are assigned higher weights. This results in a comprehensive stability assessment value, used to quantify the overall operational stability of the area where the current node is located.

[0026] Based on the above three parameters (illuminance variation rate, historical volatility, and comprehensive stability assessment value), the system sets a set of preset adjustment rules to dynamically control data acquisition and transmission behavior: Sampling frequency adjustment: If the rate of change in illumination is greater than the set threshold (e.g., 10), ), or the historical volatility index exceeds the set variance threshold (e.g., voltage variance > 0.05). The node will then initiate a sampling frequency increase strategy. The sampling period can be shortened from 60 seconds to 15 seconds; if it is below the threshold, the frequency will be gradually reduced to a minimum of 120 seconds.

[0027] Analog-to-digital conversion accuracy adjustment: When the sampling frequency is increased, in order to ensure that key features are fully captured, the sampling bit of the analog-to-digital converter (ADC) is simultaneously increased from 10 bits to 12 bits; if the sampling frequency is reduced and the system is stable, the sampling bit can be reduced to 8 bits to reduce power consumption and data volume.

[0028] Data compression ratio adjustment: The node determines the data compression ratio based on the current sampling precision. A low compression ratio (e.g., 1.5:1) is used for high precision to retain more detailed information; a high compression ratio (e.g., 4:1) is used for low precision to reduce transmission load. Compression methods can employ a combination of lossless differential coding and lightweight variable-length coding.

[0029] Transmission priority settings: Each data packet to be uploaded is packaged with an importance identifier field, calculated based on illumination fluctuation tags, abnormal records, and compression ratio. Nodes classify data into high, medium, and low priorities according to the current system load (gateway accept window size, node transmission queue length): high-priority data is immediately transmitted via the wireless link; medium-priority data is queued and waits for the channel to become available; low-priority data is delayed in uploading or temporarily cached locally for fusion processing.

[0030] To improve the ability of wireless data acquisition nodes to identify the operating status of photovoltaic modules and reduce redundant data, this invention provides a feature extraction and preliminary dimensionality reduction method based on multi-source data fusion. This method normalizes, weights, and analyzes the parameters such as voltage, current, surface temperature, and ambient light intensity collected by the nodes, ultimately extracting key operating features reflecting the dynamic behavior of the photovoltaic modules. The specific technical implementation steps are as follows: In this invention, the various physical quantities collected by the nodes have different units and large differences in dimensions. Direct feature fusion processing would cause deviations, so the data needs to be normalized.

[0031] Each data acquisition parameter (such as voltage, current, temperature, and light intensity) is assigned a corresponding theoretical maximum and minimum value, and the original values ​​are converted to a range of 0 to 1 using a linear normalization method. For example, if the component voltage range is set to 0 to 50 volts, the normalized result for the actual voltage value of 30 volts is 0.6. This normalization process unifies the scale of data across all dimensions, preventing any single physical quantity from having a dominant influence on the results during weighted calculations.

[0032] Normalized multi-source data is input into the feature fusion module. Each parameter is assigned a weighting coefficient based on its impact on the stability of the photovoltaic module's output power. For example, the current weight is set to 0.4, the voltage weight to 0.3, the surface temperature weight to 0.2, and the ambient light intensity weight to 0.1. These initial weight values ​​can be set in the factory configuration or periodically adjusted adaptively based on data collected during long-term node operation. Specifically, iterative optimization can be performed based on the correlation between output power and each input factor.

[0033] The method for calculating the comprehensive operating feature value is as follows: multiply each normalized parameter by its corresponding weight coefficient and sum them up to obtain a feature scalar representing the current node's operating status, which is denoted as the node's comprehensive feature value.

[0034] To determine the dynamic trend of photovoltaic module operating status, the node performs trend detection on this comprehensive characteristic value. The method is as follows: Set a sliding time window, the size of which can be determined based on the system sampling frequency and the most recent 10 composite feature value samples. Compare the average of the composite feature values ​​in the current window with the average of the previous window to determine the direction of change. If the current mean is higher than the mean of the previous window and increases twice consecutively, it is marked as an upward trend; If the current mean is lower than the mean of the previous window and decreases twice consecutively, it is marked as a downward trend; If the fluctuation range is lower than the set threshold (e.g., the change range is less than 0.05), it is marked as a stable trend.

[0035] Based on the trend detection results, the node outputs a set of key operating characteristics, including but not limited to the following: Power change trend: predicting the future direction of component power change based on the positive and negative correlation analysis of current and voltage trends; Temperature drift direction: identifying the continuous rise or fall trend of component surface temperature to predict component aging or local hot spots; Light response sensitivity: analyzing whether changes in light intensity trigger synchronous voltage or current fluctuations to determine the component's response performance to the external environment. This feature set is not only used for local judgment of operating status but also uploaded as an important part of the compressed local data packet to the gateway node or cloud platform for further global status analysis and intelligent maintenance decisions.

[0036] To achieve efficient classification, management, and transmission scheduling of data collected by wireless acquisition nodes in a photovoltaic module array, this invention proposes a data hierarchical mechanism based on content type and urgency. This mechanism, by hierarchically processing local characteristic data packets and combining them with a low-power wireless communication network, enables rapid response to high-priority data and energy-efficient transmission of low-priority data, effectively improving the system's real-time performance and resource utilization efficiency. The specific technical implementation steps are as follows: After the wireless data acquisition node completes local feature extraction and generates feature data packets, the system first needs to classify the data content. The classification is based on the data's source attributes and functional purpose, mainly divided into the following three categories: Real-time monitoring data refers to the characteristic values ​​of operating parameters collected and processed within the current sampling period, such as the current trend of component power change and transient surface temperature changes. Periodic statistics: refers to statistical indicators such as average, extreme values, and standard deviation generated periodically by nodes, usually based on the accumulation of a window of several minutes or hours over the past; Abnormal alarm data: refers to abnormal events detected during operation that exceed the threshold, such as sudden power drop, sudden current rise, abnormal temperature drift, etc.

[0037] Each data packet is constructed by embedding a classification identifier field within the node, indicating its category. This field can be implemented using 2-bit encoding, corresponding to the three data categories respectively. The classification information serves as the basis for subsequent priority determination and transmission scheduling.

[0038] In this invention, to achieve fine-grained control of data transmission priority, the urgency of each data packet needs to be quantitatively evaluated. The node performs the following calculation process on the key parameters contained in the characteristic data packets: Power fluctuation amplitude: defined as the difference between the component output power in the current cycle and that in the previous cycle, taking its absolute value; Temperature drift rate: the amount of change in component surface temperature per unit time, expressed in degrees Celsius per minute; Trend change direction: Different weight values ​​are assigned based on the characteristic trend judgment result (upward, downward or stable). Usually, an upward trend is assigned a value of +1, a downward trend is assigned a value of -1, and a stable trend is assigned a value of 0.

[0039] An urgency index is constructed by weighting and summing the three factors mentioned above according to their respective weight coefficients. For example, the weights are set to 0.5 for power surge, 0.3 for temperature drift, and 0.2 for trend direction. If the index value exceeds a set threshold (e.g., 0.7), the data packet is automatically marked as high-priority data. This threshold can be adjusted experimentally based on the actual system's response capability and communication capacity, or it can be dynamically issued on the platform.

[0040] Once the classification and urgency index are determined, the node allocates the data packets to three different buffer queues based on the results: High-priority queue: Used to store abnormal alarm data and data with urgency index exceeding the threshold, and adopts a real-time sending strategy; Medium priority queue: Used to store data with large status fluctuations in real-time monitoring data, with a short transmission window set, such as sending within a 30-second delay; Low-priority queue: Used to store periodic statistical data and other low-volatility data. A merge upload mechanism can be set, such as aggregating and packaging the data every 5 minutes before sending it.

[0041] To address the data loss issue caused by signal interruption, energy consumption protection, or communication conflicts at wireless acquisition nodes, and to improve the integrity and accuracy of photovoltaic array operation data, this invention introduces a missing data association and reconstruction mechanism based on a coordination coefficient at the edge gateway layer, and combines it with a lightweight graph neural network model to achieve operation status prediction and anomaly correction. Specifically, it includes the following steps: The edge gateway receives heterogeneous data streams uploaded from various wireless acquisition nodes via a low-power wireless communication module. Each data stream contains the following key fields: Node ID is used to uniquely identify the photovoltaic module that sent the data; The correlation coefficient is used to characterize the data correlation between a node and its neighboring nodes; A timestamp indicates the time when the data was collected; Local features include processed voltage, current, temperature, and illumination characteristics.

[0042] After receiving data, the gateway constructs and maintains a dynamic index table based on the node identifier, recording the location and time distribution of the N most recent valid data for each node, facilitating subsequent retrieval and clustering operations. The value of N can be set according to the sampling frequency, typically 10 to 20 samples.

[0043] If the gateway detects that a node has not uploaded data for a certain period of time (i.e., data is missing) during the indexing process, it triggers the reconstruction process. The specific method is as follows: First, the set of neighboring nodes that have a cooperative relationship with the node is found according to the index table. Nodes with a cooperative coefficient greater than a preset threshold (such as 0.7) are considered to be strongly cooperative nodes. Extract historical data (with a time error of less than 1 sampling period) that is closest to the target missing time point from these adjacent nodes to construct a time synchronization dataset; The missing data of the target node is reconstructed using a weighted interpolation fitting method on the synchronous dataset. The interpolation weight is determined by the coordination coefficient, and the higher the coordination coefficient, the greater the weight. If the missing interval exceeds 3 consecutive sampling periods, a combination of local sliding fitting and edge prediction is used for segment-level estimation.

[0044] The collaboration coefficient is a statistical correlation indicator during node operation. It can be calculated by sliding window correlation between the feature values ​​uploaded by the node and its neighboring nodes, and is updated once per hour.

[0045] The completed multi-node data is input into a lightweight graph neural network model built into the edge gateway. This model has the following characteristics: nodes correspond to photovoltaic modules; edges in the undirected graph structure represent collaborative relationships between nodes; the weights of the edges are assigned by collaboration coefficients, reflecting the strength of the correlation between nodes; the input features of each node are feature data sequences from the most recent several periods; the model integrates neighbor node information through a message passing mechanism to predict the future operating status of the target node (such as trend values ​​of power, voltage, and temperature). This graph neural network adopts a simplified GCN (Graph Convolutional Network) structure, with few parameters and low computational resource requirements, making it suitable for deployment in embedded edge computing modules.

[0046] The gateway compares the predicted values ​​output by the graph neural network with the actual values ​​reported by the nodes. When the deviation between the two exceeds a set threshold (e.g., voltage deviation exceeding 5%, temperature deviation exceeding 2℃), the data is identified as an outlier, and the following correction operations are performed: Replace the original value with the predicted value and add a "correction mark" field to record it as a model prediction intervention; If multiple adjacent nodes show abnormal deviations and the coordination coefficient drops significantly, the system records it as a "regional anomaly" and pushes it to the cloud platform for further diagnosis. All corrected data is output as structured array operation data, which includes data source, processing method identifier and correction status fields, making it easier for the upper-layer system to track the processing process.

[0047] To improve the overall data acquisition efficiency of photovoltaic module arrays, reduce node energy consumption, and enhance data timeliness, this invention introduces an acquisition strategy optimization model into the cloud platform. This model, based on intelligent analysis of structured array operating data, automatically generates control commands, which are then distributed to wireless acquisition nodes via a gateway, enabling dynamic adjustment of sampling parameters. Specifically, it includes the following steps: The cloud platform periodically receives structured array operation data uploaded by the edge gateway. This data already includes key feature values ​​and anomaly markers for each node in the current period. The platform performs the following multi-dimensional statistical processing on this data: Power fluctuation amplitude: Sliding window analysis is performed on the power value sequence of each node to extract the maximum change per unit time, which is used to measure the dynamic degree of the lighting environment in which the node is located. Data integrity rate: The ratio of the number of data packets uploaded by each node within a set period (e.g., 10 minutes) to the theoretical number of sampling times, reflecting the stability of node data; Communication load intensity: Based on the total amount of data uploaded by the node and the average bandwidth utilization rate, assess the degree of node occupancy on the current wireless network link.

[0048] The three metrics mentioned above are used to measure the necessity, stability, and transmission pressure of data collection at the nodes, and are key input parameters for subsequent sampling strategy adjustments.

[0049] Based on the above statistical indicators, the cloud platform constructs a set of data acquisition strategy optimization models. The objective function of this model is defined as a weighted combination of maximizing sampling efficiency and minimizing system energy consumption. The construction process is as follows: Sensitivity parameter setting: The power fluctuation amplitude is used as the sensitivity parameter. The greater the change amplitude, the higher the sampling priority. Constraint settings: The node data integrity rate and communication load intensity are used as dual constraints. When the integrity rate is low or the communication load is high, the sampling frequency is appropriately reduced. Priority output logic: Sort each node according to sensitivity parameters, then assign sampling levels (high, medium, low) based on constraints, and calculate the corresponding sampling granularity adjustment value (e.g., adjust the sampling period from 30 seconds to 15 seconds), feature extraction frequency coefficient (e.g., change from extracting features once per sampling to extracting features once per two samplings), and communication cycle correction amount (e.g., adjust the data upload frequency from 1 minute / time to 3 minutes / time).

[0050] The platform generates a standardized downlink control instruction set based on the parameter combination output by the optimized model. Each instruction includes: a unique identifier for the target node; an adjusted sampling granularity (in seconds); a feature extraction frequency coefficient (integer ratio); a communication upload cycle correction value (in seconds); and an effective timestamp and priority marker field.

[0051] Control commands are packaged and sent to the edge gateway via the cloud platform interface, and then forwarded by the gateway to the corresponding wireless acquisition node. Upon receiving the command, the node automatically parses it and writes it into its local control parameter area, while simultaneously sending an update status back to the gateway, thus forming a closed-loop control process.

[0052] Example 2: To verify the effectiveness of the data acquisition method for solar photovoltaic module arrays based on wireless Internet of Things described in this invention in practical applications, this invention constructed a set of experimental systems and conducted comparative experiments with traditional fixed sampling methods.

[0053] In a real photovoltaic power station, a standard array consisting of 120 photovoltaic modules was selected, and the following two data acquisition schemes were deployed: Comparison with Scheme A (traditional method): using a fixed sampling period (60 seconds / time), a fixed data upload period (1 minute / time), and no data priority division; Implementation Scheme B (Method of the Invention): A dynamic sampling mechanism is adopted, in which the sampling frequency, feature extraction frequency and communication cycle are dynamically adjusted by the cloud according to the control command issued by the strategy optimization model, and the data is transmitted according to priority classification.

[0054] Both sets of solutions ran continuously for 7 days, and the data was aggregated and uploaded to the cloud data platform through the edge gateway to record relevant operating indicators.

[0055] Performance comparison metrics and results:

[0056] In summary, the data integrity rate is calculated by the ratio of the number of data entries reported by the computing nodes to the theoretical number of samples. The implementation scheme maintains a high effective data transmission ratio even after dynamically avoiding communication congestion periods and compressing low-priority data. Node energy consumption is measured by the current acquisition module to measure the average daily power consumption of a single node. The implementation scheme significantly reduces power consumption by reducing the sampling frequency and communication frequency during low load periods. The critical event capture rate is measured by the number of times abnormal events such as simulated light surges and component temperature rises are detected and uploaded by the statistical system. The implementation scheme significantly enhances its capture capability by means of adaptive sampling granularity and priority mechanism. Network transmission load is statistically analyzed based on the total amount of data uploaded by the gateway. The implementation scheme achieves traffic compression optimization while ensuring that critical data is not lost.

[0057] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for acquiring data from a solar photovoltaic module array based on the wireless Internet of Things, characterized in that: include: Each module in the photovoltaic module array is configured with a wireless acquisition node, and the wireless acquisition node includes a unique identifier ID of the module, location information, self-learning weight parameters, and dynamic coordination factor. Each wireless acquisition node dynamically adjusts its sampling frequency and data fidelity level based on the current light intensity, historical sampling fluctuation rate, and the perception status of neighboring nodes, and performs multi-source data acquisition, including voltage, current, component surface temperature, and ambient light intensity. The collected multi-source data is processed using a weighted average and trend detection algorithm to extract key operational features. At the same time, the collaborative factors exchanged with neighboring nodes are fused to perform preliminary dimensionality reduction, generating a compressed local feature data package. The local feature data packets are divided into multiple data layers according to their content type and urgency, and are transmitted hierarchically and heterogeneously through a low-power wireless communication network. High-priority data is uploaded in real time, while low-priority data is uploaded after a delay. The edge gateway receives heterogeneous data streams from various nodes, uses the coordination coefficient to reconstruct missing data, employs a lightweight graph neural network model to predict and correct anomalies in the component's operating status, and outputs structured array operating data. The structured array's operating data is analyzed to generate an acquisition strategy optimization model. The sampling granularity, feature extraction parameters, and communication frequency of each wireless acquisition node are dynamically adjusted through downlink control commands.

2. The data acquisition method for a solar photovoltaic module array based on wireless Internet of Things according to claim 1, characterized in that: The dynamic adjustment of sampling frequency and data fidelity level includes: Measure the current light intensity and calculate the rate of change of light intensity within a short time window; Based on the local storage of the rolling sample window, the variance and peak frequency of historical sampled values ​​are statistically analyzed to generate and output the historical sampled volatility index. The system receives operational status summaries from neighboring nodes via short-range low-power wireless broadcasting, and then weights and fuses these summaries with historical volatility to obtain a comprehensive stability assessment value. Based on the rate of change of illumination, historical volatility, and comprehensive stability assessment value, the sampling frequency, analog-to-digital conversion sampling accuracy, data compression ratio, and transmission priority are dynamically adjusted according to preset rules.

3. The data acquisition method for a solar photovoltaic module array based on wireless Internet of Things according to claim 2, characterized in that: The dynamic adjustment of sampling frequency, analog-to-digital conversion sampling accuracy, data compression ratio, and transmission priority includes: When the rate of change of illumination is greater than the set threshold, the sampling frequency is automatically increased; otherwise, the sampling frequency is decreased. Based on the adjusted sampling frequency, the sampling accuracy is improved when the sampling frequency is increased; Based on the current analog-to-digital conversion accuracy, the nodes dynamically adjust the data compression parameters; Based on the importance and load of the compressed data packets, high, medium, and low transmission priorities are assigned to the data packets, with high-priority data being uploaded via the wireless link first.

4. The data acquisition method for a solar photovoltaic module array based on wireless Internet of Things according to claim 3, characterized in that: The process of extracting key operational features from the collected multi-source data using weighted averaging and trend detection algorithms includes: The collected voltage, current, component surface temperature and ambient light intensity data are subjected to dimension unification and normalization processing. Based on the degree of influence of each parameter on the output power stability of the photovoltaic module, a weight coefficient is assigned to each parameter, and the comprehensive operating characteristic value of the node is calculated. Within a set sliding time window, trend analysis is performed on continuous comprehensive characteristic values. By comparing the changing direction of the current characteristic value with the average value of the previous window, the upward, downward, or stable trend states can be identified. Based on the trend detection results, key operational feature sets are extracted, including power change trends, temperature drift direction, and light response sensitivity.

5. The data acquisition method for a solar photovoltaic module array based on wireless Internet of Things according to claim 4, characterized in that: The local feature data packets are divided into multiple data layers based on content type and urgency, including: When generating local feature data packets, classification labels are set according to the data source attributes and content types, which include three types of labels: real-time monitoring data, periodic statistical data, and abnormal alarm data. Based on the power fluctuation magnitude, temperature drift rate, and trend change direction contained in the data packet, an urgency index is calculated. When the index exceeds a set threshold, the data is marked as high priority. Data packets are stored in high, medium, and low priority buffer queues according to their classification identifier and urgency level. The high priority queue adopts a real-time sending strategy, while the medium and low priority queues enter the delayed aggregation waiting area.

6. The data acquisition method for a solar photovoltaic module array based on wireless Internet of Things according to claim 5, characterized in that: The edge gateway receives heterogeneous data streams from various nodes and uses a coordination coefficient to reconstruct missing data, including: Receive heterogeneous data streams containing node identifiers, coordination coefficients, and timestamps from multiple wireless acquisition nodes, and build an index table based on the node identifiers; When a missing data node is detected, the missing data segment is reconstructed by interpolation fitting based on the coordination coefficient between the node and its neighboring nodes, selecting a dataset that is time-synchronized and highly coordinated from the historical data of the neighboring nodes. The reconstructed multi-node data is input into a lightweight graph neural network model, and the edge weights between nodes are defined by the synergy coefficient. Compare the predicted value with the actual collected value. When the deviation exceeds the set threshold, perform anomaly correction and mark the source of correction.

7. The data acquisition method for a solar photovoltaic module array based on wireless Internet of Things according to claim 6, characterized in that: The structured array's operational data is analyzed to generate an optimization model for the acquisition strategy, including: Perform multi-dimensional statistical analysis on the structured array operation data uploaded by the edge gateway to extract power fluctuation amplitude, node data integrity rate and communication load intensity; Based on the analysis indicators, an optimization model for the acquisition strategy is established. The model uses the power fluctuation amplitude as a sensitive parameter and node energy consumption and communication load as constraints to output the sampling priority and parameter adjustment suggestions for each node. A downlink control instruction set is generated based on the optimization model results, including sampling granularity adjustment values, feature extraction frequency coefficients, and communication cycle correction amounts.

Images