Photovoltaic power station equipment operation data analysis method and system based on Internet of Things

Abstract

The invention relates to the technical field of intelligent operation and maintenance of photovoltaic power station equipment, and discloses a photovoltaic power station equipment operation data analysis method and system based on the Internet of Things. The operation data analysis method is applied to data analysis equipment and specifically comprises the following steps that S101, a data analysis request is received, and the data analysis request collects parameter data of current, voltage, temperature, irradiance and environment humidity in real time through Internet of Things sensor nodes deployed on a photovoltaic module, a combiner box, an inverter and a meteorological station; according to the method, the problem of space-time misalignment of multi-source heterogeneous data is solved, the outlier recognition accuracy is improved, the data availability rate is greatly optimized, the limitation of traditional single-dimensional analysis is broken through by innovative four-dimensional feature engineering, the ripple spectrum entropy in the electrical features and the response delay time in the environment coupling features are effectively improved, and the reliability of the multi-source heterogeneous data is improved. And by combining a thermal-electric propagation model constructed by GCN, the detection rate of early faults such as microscopic subfissure is increased from 60% to 95%.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent operation and maintenance of photovoltaic power plant equipment, and more specifically, to a method and system for analyzing the operation data of photovoltaic power plant equipment based on the Internet of Things. Background Technology

[0002] With the continuous growth of global energy demand and the increasing prominence of environmental issues, the development and utilization of clean energy has become a focus of attention for all countries. As an important renewable energy source, photovoltaic power generation has been widely used and developed rapidly due to its advantages such as being clean, environmentally friendly, and rich in resources. In particular, distributed photovoltaic power stations, due to their flexible deployment methods and grid-friendly nature, have gradually become an important part of the photovoltaic power generation field.

[0003] Due to the dispersed and diverse nature of equipment and the complex operating environment, fault diagnosis and maintenance of distributed photovoltaic (PV) power plants are particularly important. Traditional PV power plant equipment diagnosis methods mainly rely on regular manual inspections and simple monitoring systems. Currently, PV power plants generally adopt a hierarchical monitoring architecture: at the data acquisition layer, module strings, inverters, and weather sensors are connected through industrial protocols such as Modbus and CAN bus, and preliminary data integration is achieved using protocol conversion gateways; at the fault detection layer, static threshold alarms are mainly set using SCADA systems, supplemented by quarterly manual infrared inspections; at the optimization control layer, fixed cleaning plans are formulated based on historical power generation curves, and basic MPPT tracking is achieved through string capacity matching design.

[0004] Traditional photovoltaic power plant operation data analysis lacks systematicity: First, the fusion of multi-source heterogeneous data failed, the timestamp deviation of data from different protocol devices exceeded ±30 seconds, electromagnetic interference reduced the signal-to-noise ratio of current signals to as low as 15dB, and traditional Kalman filtering did not consider the nonlinear drift of sensors, resulting in a 12% distortion rate in the preprocessed data. Secondly, the depth of fault diagnosis is insufficient; static threshold alarms are completely ineffective for early latent defects (such as the initial stage of PID effect); the SVM model has a high misjudgment rate of up to 35% for hot spots and dust obstruction due to weak feature engineering (using only 5-dimensional features such as current and voltage); manual inspection has a long cycle and a 40% failure rate for micro-cracks. Third, the control strategy is rigid, the fixed cleaning plan does not incorporate meteorological forecast data, the invalid cleaning frequency accounts for 30%, the string reconstruction relies on the initial design and cannot respond to dynamic shadow changes, resulting in a large annual power generation loss. Moreover, data security and traceability are lacking, the transmission from the gateway to the cloud platform uses the plaintext MQTT protocol, and 23% of the security incidents in the industry in the past three years were data tampering attacks.

[0005] Based on the above, there is an urgent need to design a method and system for analyzing the operation data of photovoltaic power station equipment based on the Internet of Things, so as to solve the technical problems mentioned above. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for analyzing the operation data of photovoltaic power station equipment based on the Internet of Things, which significantly improves the data governance and fault diagnosis capabilities of photovoltaic power stations and greatly optimizes the data availability. Secondly, the innovative four-dimensional feature engineering breaks through the limitations of traditional single-dimensional analysis, and the ripple spectrum entropy value in electrical features and the response delay time in environmental coupling features are intended to solve the problems in the existing technology.

[0007] This invention is implemented as follows: a data analysis method for photovoltaic power station equipment operation based on the Internet of Things, applied to data analysis equipment, specifically including the following steps: S101: Receive a data analysis request, which collects parameter data such as current, voltage, temperature, irradiance, and ambient humidity in real time through IoT sensor nodes deployed on photovoltaic modules, combiner boxes, inverters, and weather stations. S102: Perform three-level verification on the parameter data. First, remove the 3σ outliers of current and voltage based on the Laida criterion. Second, compensate for missing data segments caused by network delay through linear interpolation. Finally, use wavelet transform to filter out high-frequency electromagnetic interference noise and compensate for sensor data drift to obtain the preprocessed data for three-level verification. S103: Based on preprocessed data, start the distributed feature computing engine, extract four-dimensional feature sets by device ID, generate device ID associated feature vectors to build InfluxDB storage feature library, and simultaneously configure downsampling continuous query and hot and cold data stratification strategies to achieve tens of thousands of data points written per second and 50ms query response. S104: A hierarchical diagnostic model is used to analyze the device ID associated feature vector. The first layer, random forest, calls electrical features and time sequence features to identify basic faults. The second layer, GCN, fuses infrared thermal imaging with temperature gradient field / spectral entropy values ​​in the feature library to construct a heat-electric propagation model to locate fault points and evaluate dynamic weighted efficiency decay rate, temperature rise anomaly and insulation degradation trend. The output is fault code and life prediction value based on device ID associated feature vector. S105: Based on the output fault codes and life prediction values, the string dynamic reconstruction algorithm is started to replan the MPPT tracking path. For strings with detected PID risks, nighttime reverse bias repair is automatically triggered. At the same time, the equipment health status is mapped to the power plant power prediction model. AGC / AVC control commands are rolled over with a preset time granularity and sent to the inverter cluster through the OPCUA protocol to complete the phased photovoltaic power plant equipment operation data analysis and correction.

[0008] Further, in S101, a data analysis request is received, the data analysis request being processed by IoT sensor nodes deployed on photovoltaic modules, combiner boxes, inverters, and weather stations, including: The sensor nodes use a hybrid network of LoRaWAN protocol and ZigBee to synchronously transmit data to the edge computing gateway at a 5-minute interval. The edge computing gateway timestamps and binds data to device IDs, generates structured data packets, and encrypts and uploads them to the cloud platform. It also integrates the switch status signals of the SCADA system and power grid dispatch instructions to build a dynamic database containing device physical parameters, environmental parameters, and power grid interaction parameters.

[0009] Furthermore, in S102, a three-level verification is performed on the parameter data. First, outliers of current and voltage (3σ) are removed based on the Raida criterion. Second, missing data segments caused by network latency are compensated through linear interpolation, including: A dynamic sliding window is constructed by grouping devices by ID. The window is set to 60 minutes in sunny weather and automatically shortened to 15 minutes in cloudy or rainy weather to capture sudden changes. The mean μ and standard deviation σ of current / voltage within the window are calculated in real time, and data points exceeding μ±3σ are marked as outliers. When the outlier rate of a device exceeds 20%, compare the data with the data of devices in the same string. If the difference is less than 5%, it is determined to be a sensor failure and a calibration alarm is triggered. For data missing segments lasting less than 5 seconds, a time-weighted linear interpolation method is used for compensation. When the data missing duration exceeds 5 seconds, a device collaborative compensation mechanism is activated, and a source flag is added to all compensated data.

[0010] Furthermore, for data missing segments lasting less than 5 seconds, a time-weighted linear interpolation method is used for compensation. When the data missing duration exceeds 5 seconds, a device collaborative compensation mechanism is activated, including: For data missing segments lasting less than 5 seconds, based on the relative position of the current time within the missing time period, the starting and ending values ​​are linearly weighted according to the time ratio to obtain compensation data for a smooth transition. When the data loss time exceeds 5 seconds, select 3 adjacent devices of the same model in physical location, collect their monitoring data at the corresponding time, calculate the compensation value using the distance inverse weighting strategy, and use the weighted average of the three devices as the compensation data for the missing device. The original collected data is marked as 0, the short-term interpolated data is marked as 1, and the long-term collaborative compensation data is marked as 2. This flag will be passed to the subsequent data analysis and the differentiated processing of data credibility.

[0011] Furthermore, wavelet transform is used to filter out high-frequency electromagnetic interference noise and compensate for sensor data drift, resulting in preprocessed data for three-level verification, including: The Daubechies5 wavelet basis is used to decompose the current and voltage signals into 5 levels. The high-frequency detail coefficients of levels 1-3 are thoroughly filtered out by the hard thresholding method to remove inverter switching noise >20kHz. The mid-frequency coefficients of levels 4-5 are decomposed by the soft thresholding method to retain potential fault characteristics. To compensate for sensor data drift, an adaptive calibration model based on historical temperature rise curves is established. The compensation coefficient is dynamically adjusted using ambient humidity as a correction factor. A terrain shadow compensation algorithm is implemented for irradiance data, and the actual amount of light received on the component surface is calculated in conjunction with the three-dimensional model of the power station.

[0012] Further, in S103, a distributed feature computing engine is started based on the preprocessed data to extract a four-dimensional feature set by device ID. The four-dimensional feature set includes: Electrical characteristics, including the inflection point slope of the IV curve of the module string, the inverter conversion efficiency deviation, and the DC side ripple coefficient; Thermal characteristics, including the temperature rise gradient of the module backsheet and the temperature dispersion coefficient between strings; Environmental coupling characteristics, including irradiance-power output response delay time, humidity and insulation resistance Pearson correlation coefficient; Time series characteristics include the sliding window variance of power decay rate and the nighttime reverse current abrupt change of microcracked cells.

[0013] Furthermore, in S104, a heat-electric propagation model is constructed to locate fault points, assess the dynamic weighted efficiency decay rate, temperature rise anomaly, and insulation degradation trend, and output fault codes and lifetime prediction values ​​based on equipment ID-associated feature vectors, including: Using the unique device ID of the photovoltaic module as the node identifier, a topology network is constructed based on the electrical wiring connection relationship and the heat radiation transfer path. Each node loads electrical characteristics and real-time infrared thermal imaging temperature distribution data from the InfluxDB storage feature library. The weights are calculated based on the electrical correlation strength and heat exchange efficiency between adjacent nodes. The features of neighboring nodes are aggregated layer by layer and their own states are updated to generate a full-field fault probability distribution map, which can accurately locate the location of anomalies.

[0014] Furthermore, in S105, based on the output fault code and lifetime prediction value, the string dynamic reconstruction algorithm is initiated to replan the MPPT tracking path. For strings with detected PID risk, nighttime reverse bias repair is automatically triggered, including: Based on the fault code, locate the abnormal component ID, avoid components in the shadow / hot spot area, re-constrain the string electrical topology, and re-plan the string electrical topology. The constraints include the upper limit of single string voltage, the current imbalance between strings and the physical wiring distance limit. Relay switch matrix instructions are generated and reconstructed through the smart junction box. Differentiated repairs are implemented based on relay switch matrix instruction reconstruction. For low-to-medium risk, a fixed bias voltage of -200V / 2 hours is applied, while for high-risk, dynamic bias voltage adjustment and intelligent duration control are adopted. The optimal resistance value is dynamically calculated by combining the insulation resistance decay rate of the InfluxDB storage feature library. The leakage current change is monitored in real time during the repair process, and the operation is automatically executed at night. After the repair, the verification data is sent back to update the equipment health record.

[0015] Compared with existing technologies, the IoT-based photovoltaic power plant equipment operation data analysis method and system provided by this invention have the following advantages: 1. Significantly improves the data governance and fault diagnosis capabilities of photovoltaic power plants. Through hybrid networking and dynamic sliding window verification mechanisms, it solves the problem of spatiotemporal inaccuracy of multi-source heterogeneous data, improves the accuracy of outlier identification, and greatly optimizes data availability. Secondly, the innovative four-dimensional feature engineering breaks through the limitations of traditional single-dimensional analysis. The ripple spectrum entropy value in electrical features and the response delay time in environmental coupling features, combined with the heat-electric propagation model constructed by GCN, increase the detection rate of early faults such as microcracks from 60% to 95%, with high positioning accuracy and reduced fault misjudgment rate. 2. Achieving closed-loop optimization and system gain, the response latency of the string dynamic reconstruction algorithm is compressed to 200ms, reducing power generation losses in shading scenarios. The PID graded repair mechanism combined with dynamic voltage adjustment based on insulation resistance attenuation rate results in high repair success rate and reduced energy consumption. The AGC / AVC rolling correction driven by equipment health reduces power prediction error. Blockchain notarization ensures decision traceability and reduces maintenance costs, greatly improving power plant availability and increasing annual power generation gains. The verifiable carbon footprint throughout the entire life cycle supports green finance applications.

[0016] A photovoltaic power plant equipment operation data analysis system based on the Internet of Things is used to execute the above-mentioned photovoltaic power plant equipment operation data analysis method. The data analysis system includes: The multi-source data acquisition module is used to collect current, voltage, temperature, irradiance, and ambient humidity data in real time through sensor nodes of photovoltaic modules / combiner boxes / inverters / weather stations, and supports hybrid networking transmission of LoRaWAN and ZigBee; The three-level verification module is used to perform outlier removal by the Laida criterion, linear interpolation data compensation, wavelet denoising and temperature drift calibration, and output preprocessed data. The feature calculation engine module is used to extract electrical / thermal / environmental / temporal four-dimensional feature sets in parallel by device ID, build the InfluxDB time series database, and implement hierarchical management of hot and cold data. The hierarchical diagnostic module is used to identify basic faults using random forest, locate fault points by fusing infrared data with GCN, dynamically calculate the degradation index and output the lifetime prediction value. The closed-loop control module is used to perform string dynamic reconstruction, PID reverse bias repair, AGC / AVC instruction rolling optimization, and to store decision logs through blockchain.

[0017] Specifically, the feature calculation engine module includes: Electrical characteristic unit, used to calculate the inflection point slope of IV curve and ripple spectrum entropy value; Thermal feature units are used to generate component temperature gradient field distribution maps. An environmental coupling unit is used to construct the irradiance-power response transfer function; The timing feature unit is used to extract the frequency domain abrupt change of the reverse current at night. The database construction unit is used to process and merge the acquired four-dimensional feature set to build the InfluxDB time series database and implement hierarchical management of hot and cold data. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the method for analyzing the operation data of photovoltaic power station equipment based on the Internet of Things proposed in this invention. Figure 2 This is a flowchart illustrating the three-level verification process for parameter data in the IoT-based photovoltaic power plant equipment operation data analysis method proposed in this invention. Figure 3 This is a schematic diagram of the structure of the photovoltaic power station equipment operation data analysis system based on the Internet of Things proposed in this invention; Figure 4 This is a schematic diagram of the feature calculation engine module in the IoT-based photovoltaic power plant equipment operation data analysis system proposed in this invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] The implementation of the present invention will be described in detail below with reference to specific embodiments.

[0021] In the accompanying drawings of this embodiment, the same or similar reference numerals correspond to the same or similar components. In the description of this invention, it should be understood that if terms such as "upper," "lower," "left," and "right" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting this invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0022] Reference Figure 1-2 As shown, the IoT-based photovoltaic power plant equipment operation data analysis method, applied to data analysis equipment, specifically includes the following steps: S101: Receives data analysis requests. The data analysis requests collect parameter data such as current, voltage, temperature, irradiance, and ambient humidity in real time through IoT sensor nodes deployed on photovoltaic modules, combiner boxes, inverters, and weather stations. This includes receiving data analysis requests, which are processed through IoT sensor nodes deployed on photovoltaic modules, combiner boxes, inverters, and weather stations, including: The sensor nodes use a hybrid network of LoRaWAN protocol and ZigBee to synchronously transmit data to the edge computing gateway at a 5-minute interval. The edge computing gateway timestamps and binds data to device IDs, generates structured data packets, and encrypts and uploads them to the cloud platform. It also integrates the switch status signals of the SCADA system and power grid dispatch instructions to build a dynamic database containing device physical parameters, environmental parameters, and power grid interaction parameters. S102: Perform three-level verification on the parameter data. First, remove the 3σ outliers of current and voltage based on the Laida criterion. Second, compensate for missing data segments caused by network delay through linear interpolation. Finally, use wavelet transform to filter out high-frequency electromagnetic interference noise and compensate for sensor data drift to obtain the preprocessed data for three-level verification. The parameter data undergoes a three-level verification process. First, outliers in current and voltage are removed based on the Laida criterion. Second, missing data segments caused by network latency are compensated using linear interpolation, including: A dynamic sliding window is constructed by grouping devices by ID. The window is set to 60 minutes in sunny weather and automatically shortened to 15 minutes in cloudy or rainy weather to capture sudden changes. The mean μ and standard deviation σ of current / voltage within the window are calculated in real time, and data points exceeding μ±3σ are marked as outliers. When the outlier rate of a device exceeds 20%, compare the data with the data of devices in the same string. If the difference is less than 5%, it is determined to be a sensor failure and a calibration alarm is triggered. For data missing segments lasting less than 5 seconds, a time-weighted linear interpolation method is used for compensation. When the data missing duration exceeds 5 seconds, a device collaborative compensation mechanism is activated, and a source flag is added to all compensated data. S103: Based on preprocessed data, start the distributed feature computing engine, extract four-dimensional feature sets by device ID, generate device ID associated feature vectors to build InfluxDB storage feature library, and simultaneously configure downsampling continuous query and hot and cold data stratification strategies to achieve tens of thousands of data points written per second and 50ms query response. Specifically, a distributed feature computation engine is launched based on preprocessed data to extract a four-dimensional feature set by device ID. The four-dimensional feature set includes: Electrical characteristics, including the inflection point slope of the IV curve of the module string, the inverter conversion efficiency deviation, and the DC side ripple coefficient; Thermal characteristics, including the temperature rise gradient of the module backsheet and the temperature dispersion coefficient between strings; Environmental coupling characteristics, including irradiance-power output response delay time, humidity and insulation resistance Pearson correlation coefficient; Time series characteristics, including the sliding window variance of power decay rate and the nighttime reverse current abrupt change of microcracked cells; S104: A hierarchical diagnostic model is used to analyze the device ID associated feature vector. The first layer, random forest, calls electrical features and time sequence features to identify basic faults. The second layer, GCN, fuses infrared thermal imaging with temperature gradient field / spectral entropy values ​​in the feature library to construct a heat-electric propagation model to locate fault points and evaluate dynamic weighted efficiency decay rate, temperature rise anomaly and insulation degradation trend. The output is fault code and life prediction value based on device ID associated feature vector. The process involves constructing a heat-electric propagation model to locate fault points, assessing dynamically weighted efficiency decay rate, temperature rise anomaly, and insulation degradation trend, and outputting fault codes and lifetime predictions based on device ID-associated feature vectors. Using the unique device ID of the photovoltaic module as the node identifier, a topology network is constructed based on the electrical wiring connection relationship and the heat radiation transfer path. Each node loads electrical characteristics and real-time infrared thermal imaging temperature distribution data from the InfluxDB storage feature library. The weights are calculated based on the electrical correlation strength and heat exchange efficiency between adjacent nodes. The features of neighboring nodes are aggregated layer by layer and their own states are updated to generate a full-field fault probability distribution map and accurately locate the location of anomalies. S105: Based on the output fault codes and life prediction values, the string dynamic reconstruction algorithm is initiated to replan the MPPT tracking path. For strings with detected PID risks, nighttime reverse bias repair is automatically triggered. At the same time, the equipment health status is mapped to the power plant power prediction model. AGC / AVC control commands are rolled and corrected at preset time granularity and sent to the inverter cluster through the OPCUA protocol. This completes the phased analysis and correction of photovoltaic power plant equipment operation data, which significantly improves the data governance and fault diagnosis capabilities of photovoltaic power plants. Through hybrid networking and dynamic sliding window verification mechanism, the problem of spatiotemporal inaccuracy of multi-source heterogeneous data is solved, the accuracy of outlier identification is improved, and the data availability is greatly optimized. Secondly, the innovative four-dimensional feature engineering breaks through the limitations of traditional single-dimensional analysis. The ripple spectrum entropy value in electrical features and the response delay time in environmental coupling features, combined with the heat-electric propagation model constructed by GCN, increase the detection rate of early faults such as microcracks from 60% to 95%, with high positioning accuracy and reduced fault misjudgment rate.

[0023] In S102 of this embodiment, for data missing segments lasting less than 5 seconds, a time-weighted linear interpolation method is used for compensation. When the data missing duration exceeds 5 seconds, a device collaborative compensation mechanism is activated, including: For data missing segments lasting less than 5 seconds, based on the relative position of the current time within the missing time period, the starting and ending values ​​are linearly weighted according to the time ratio to obtain compensation data for a smooth transition. When the data loss time exceeds 5 seconds, select 3 adjacent devices of the same model in physical location, collect their monitoring data at the corresponding time, calculate the compensation value using the distance inverse weighting strategy, and use the weighted average of the three devices as the compensation data for the missing device. The original collected data is marked as 0, the short-term interpolated data is marked as 1, and the long-term collaborative compensation data is marked as 2. This flag will be passed to the subsequent data analysis and the differentiated processing of data credibility.

[0024] In this embodiment, wavelet transform is used to filter out high-frequency electromagnetic interference noise and compensate for sensor data drift, resulting in preprocessed data for three-level verification, including: The Daubechies5 wavelet basis is used to decompose the current and voltage signals into 5 levels. The high-frequency detail coefficients of levels 1-3 are thoroughly filtered out by the hard thresholding method to remove inverter switching noise >20kHz. The mid-frequency coefficients of levels 4-5 are decomposed by the soft thresholding method to retain potential fault characteristics. To compensate for sensor data drift, an adaptive calibration model based on historical temperature rise curves is established. The compensation coefficient is dynamically adjusted using ambient humidity as a correction factor. A terrain shadow compensation algorithm is implemented for irradiance data, and the actual amount of light received on the component surface is calculated in conjunction with the three-dimensional model of the power station.

[0025] In S105 of this embodiment, based on the output fault code and lifetime prediction value, the string dynamic reconstruction algorithm is activated to replan the MPPT tracking path. For strings with detected PID risk, nighttime reverse bias repair is automatically triggered, including: Based on the fault code, locate the abnormal component ID, avoid components in the shadow / hot spot area, re-constrain the string electrical topology, and re-plan the string electrical topology. The constraints include the upper limit of single string voltage, the current imbalance between strings and the physical wiring distance limit. Relay switch matrix instructions are generated and reconstructed through the smart junction box. Differentiated repairs are implemented based on relay switch matrix instruction reconstruction. For low-to-medium risk, a fixed bias voltage of -200V / 2 hours is applied, while for high-risk, dynamic bias voltage adjustment and intelligent duration control are adopted. The optimal resistance value is dynamically calculated by combining the insulation resistance decay rate of the InfluxDB storage feature library. The leakage current change is monitored in real time during the repair process, and the operation is automatically executed at night. After the repair, the verification data is sent back to update the equipment health record.

[0026] This technical solution achieves closed-loop optimization and system gain. The response latency of the string dynamic reconstruction algorithm is compressed to 200ms, reducing power generation losses in shading scenarios. The PID graded repair mechanism combined with dynamic voltage adjustment based on insulation resistance attenuation rate results in high repair success rate and reduced energy consumption. The AGC / AVC rolling correction driven by equipment health reduces power prediction error. Blockchain notarization ensures decision traceability and reduces maintenance costs, greatly improving power plant availability and increasing annual power generation gain. The verifiable carbon footprint throughout the entire life cycle supports green finance applications.

[0027] Reference Figure 3-4As shown, the IoT-based photovoltaic power station equipment operation data analysis system is used to execute the aforementioned photovoltaic power station equipment operation data analysis method. The data analysis system includes: a multi-source data acquisition module, used to collect real-time current, voltage, temperature, irradiance, and ambient humidity data through sensor nodes of photovoltaic modules / combiner boxes / inverters / weather stations, supporting LoRaWAN and ZigBee hybrid networking transmission; a three-level verification module, used to perform Raida criterion outlier removal, linear interpolation data compensation, wavelet denoising, and temperature drift calibration, outputting preprocessed data; a feature calculation engine module, used to extract electrical / thermal / environmental / time-series four-dimensional feature sets in parallel according to device ID, construct an InfluxDB time-series database, and implement hierarchical management of hot and cold data; and a hierarchical diagnostic... The fault identification module uses random forests to identify basic faults, integrates GCN with infrared data to locate fault points, dynamically calculates the degradation index, and outputs a lifetime prediction value. The closed-loop control module performs string dynamic reconstruction, PID reverse bias repair, and AGC / AVC instruction rolling optimization. It uses blockchain to store decision logs and employs a hybrid networking and dynamic sliding window verification mechanism to address the spatiotemporal inaccuracies of multi-source heterogeneous data, improving outlier identification accuracy and significantly optimizing data availability. Furthermore, the innovative four-dimensional feature engineering breaks through the limitations of traditional single-dimensional analysis. The ripple spectrum entropy value in electrical features and the response delay time in environmental coupling features, combined with the heat-electric propagation model constructed by GCN, increase the detection rate of early faults such as microcracks from 60% to 95%. In this embodiment, the feature calculation engine module includes: an electrical feature unit for calculating the inflection point slope of the IV curve and the ripple spectrum entropy value; a thermal feature unit for generating a component temperature gradient field distribution map; an environmental coupling unit for constructing an irradiance-power response transfer function; a time-series feature unit for extracting the frequency domain mutation of the reverse current at night; and a database construction unit for processing and merging the acquired four-dimensional feature set, constructing an InfluxDB time-series database, and implementing hierarchical management of hot and cold data to achieve closed-loop optimization and system gain. The response delay of the string dynamic reconstruction algorithm is compressed to 200ms, reducing power generation loss in shading scenarios. The PID graded repair mechanism combined with dynamic voltage adjustment based on insulation resistance attenuation rate results in high repair success rate and reduced energy consumption. The AGC / AVC rolling correction driven by equipment health reduces power prediction error.

[0028] This technical solution significantly enhances the data governance and fault diagnosis capabilities of photovoltaic power plants. Through hybrid networking and dynamic sliding window verification mechanisms, it solves the problem of spatiotemporal inaccuracies in multi-source heterogeneous data, improves the accuracy of outlier identification, and greatly optimizes data availability. Secondly, the innovative four-dimensional feature engineering breaks through the limitations of traditional single-dimensional analysis. The ripple spectrum entropy value in electrical features and the response delay time in environmental coupling features, combined with the heat-electric propagation model constructed by GCN, increase the detection rate of early faults such as microcracks from 60% to 95%, improves positioning accuracy, reduces the fault misjudgment rate, and ensures decision traceability by blockchain evidence storage. The cost of dimensional analysis is reduced, the availability of power plants is greatly improved, the annual power generation gain is increased, and the carbon footprint throughout the entire life cycle can be verified to support green finance applications.

[0029] In this embodiment, the entire operation process can be controlled by a computer to provide signal feedback and implement the steps sequentially. These are all conventional knowledge in current automation control, and will not be elaborated on in this embodiment.

[0030] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for analyzing the operation data of photovoltaic power station equipment based on the Internet of Things, characterized in that, When applied to data analysis equipment, the specific steps include: S101: Receive a data analysis request, which collects parameter data such as current, voltage, temperature, irradiance, and ambient humidity in real time through IoT sensor nodes deployed on photovoltaic modules, combiner boxes, inverters, and weather stations. S102: Perform three-level verification on the parameter data. First, remove the 3σ outliers of current and voltage based on the Laida criterion. Second, compensate for missing data segments caused by network delay through linear interpolation. Finally, use wavelet transform to filter out high-frequency electromagnetic interference noise and compensate for sensor data drift to obtain the preprocessed data for three-level verification. S103: Based on preprocessed data, start the distributed feature computing engine, extract four-dimensional feature sets by device ID, generate device ID associated feature vectors to build InfluxDB storage feature library, and simultaneously configure downsampling continuous query and hot and cold data stratification strategies to achieve tens of thousands of data points written per second and 50ms query response. S104: A hierarchical diagnostic model is used to analyze the device ID associated feature vector. The first layer, random forest, calls electrical features and time sequence features to identify basic faults. The second layer, GCN, fuses infrared thermal imaging with temperature gradient field / spectral entropy values ​​in the feature library to construct a heat-electric propagation model to locate fault points and evaluate dynamic weighted efficiency decay rate, temperature rise anomaly and insulation degradation trend. The output is fault code and life prediction value based on device ID associated feature vector. S105: Based on the output fault codes and life prediction values, the string dynamic reconstruction algorithm is started to replan the MPPT tracking path. For strings with detected PID risks, nighttime reverse bias repair is automatically triggered. At the same time, the equipment health status is mapped to the power plant power prediction model. AGC / AVC control commands are rolled over with a preset time granularity and sent to the inverter cluster through the OPCUA protocol to complete the phased photovoltaic power plant equipment operation data analysis and correction.

2. The method for analyzing photovoltaic power station equipment operation data based on the Internet of Things as described in claim 1, characterized in that, In S101, a data analysis request is received. This data analysis request is processed via IoT sensor nodes deployed on photovoltaic modules, combiner boxes, inverters, and weather stations, including: The sensor nodes use a hybrid network of LoRaWAN protocol and ZigBee to synchronously transmit data to the edge computing gateway at a 5-minute interval. The edge computing gateway timestamps and binds data to device IDs, generates structured data packets, and encrypts and uploads them to the cloud platform. It also integrates the switch status signals of the SCADA system and power grid dispatch instructions to build a dynamic database containing device physical parameters, environmental parameters, and power grid interaction parameters.

3. The method for analyzing photovoltaic power station equipment operation data based on the Internet of Things as described in claim 2, characterized in that, In S102, a three-level verification is performed on the parameter data. First, outliers of current and voltage (3σ) are removed based on the Laida criterion. Second, missing data segments caused by network latency are compensated through linear interpolation, including: A dynamic sliding window is constructed by grouping devices by ID. The window is set to 60 minutes in sunny weather and automatically shortened to 15 minutes in cloudy or rainy weather to capture sudden changes. The mean μ and standard deviation σ of current / voltage within the window are calculated in real time, and data points exceeding μ±3σ are marked as outliers. When the outlier ratio of a device exceeds 20%, compare the data with the data of devices in the same string. If the difference is less than 5%, it is determined to be a sensor fault and a calibration alarm is triggered. For data missing segments lasting less than 5 seconds, a time-weighted linear interpolation method is used for compensation. When the data missing duration exceeds 5 seconds, a device collaborative compensation mechanism is activated, and a source flag is added to all compensated data.

4. The method for analyzing photovoltaic power station equipment operation data based on the Internet of Things as described in claim 3, characterized in that, For data missing segments lasting less than 5 seconds, a time-weighted linear interpolation method is used for compensation. When the data missing duration exceeds 5 seconds, a device collaborative compensation mechanism is activated, including: For data missing segments lasting less than 5 seconds, based on the current position within the missing time period, the starting and ending values ​​are linearly weighted according to the time ratio to obtain compensation data for a smooth transition. When the data loss time exceeds 5 seconds, select 3 adjacent devices of the same model in physical location, collect their monitoring data at the corresponding time, calculate the compensation value using the distance inverse weighting strategy, and use the weighted average of the three devices as the compensation data for the missing device. The original collected data is marked as 0, the short-term interpolated data is marked as 1, and the long-term collaborative compensation data is marked as 2. This flag will be passed to the subsequent data analysis and the differentiated processing of data credibility.

5. The method for analyzing photovoltaic power station equipment operation data based on the Internet of Things as described in claim 4, characterized in that, Wavelet transform is used to filter out high-frequency electromagnetic interference noise and compensate for sensor data drift, resulting in preprocessed data for three-level verification, including: The Daubechies5 wavelet basis is used to decompose the current and voltage signals into 5 layers. The high-frequency detail coefficients of layers 1-3 are thoroughly filtered out by the hard thresholding method to remove inverter switching noise >20kHz. The mid-frequency coefficients of layers 4-5 are decomposed by the soft thresholding method to retain potential fault characteristics. To compensate for sensor data drift, an adaptive calibration model based on historical temperature rise curves is established. The compensation coefficient is dynamically adjusted using ambient humidity as a correction factor. A terrain shadow compensation algorithm is implemented for irradiance data, and the actual amount of light received on the component surface is calculated in conjunction with the three-dimensional model of the power station.

6. The method for analyzing photovoltaic power station equipment operation data based on the Internet of Things as described in claim 5, characterized in that, In S103, a distributed feature computing engine is started based on the preprocessed data to extract a four-dimensional feature set by device ID. The four-dimensional feature set includes: Electrical characteristics, including the inflection point slope of the IV curve of the module string, the inverter conversion efficiency deviation, and the DC side ripple coefficient; Thermal characteristics, including the temperature rise gradient of the module backsheet and the temperature dispersion coefficient between strings; Environmental coupling characteristics, including irradiance-power output response delay time, humidity and insulation resistance Pearson correlation coefficient; Time series characteristics include the sliding window variance of power decay rate and the nighttime reverse current abrupt change of microcracked cells.

7. The method for analyzing photovoltaic power station equipment operation data based on the Internet of Things as described in claim 6, characterized in that, In S104, a heat-electric propagation model is constructed to locate fault points, assess dynamically weighted efficiency decay rate, temperature rise anomaly, and insulation degradation trend, and output fault codes and lifetime predictions based on device ID-associated feature vectors, including: Using the unique device ID of the photovoltaic module as the node identifier, a topology network is constructed based on the electrical wiring connection relationship and the heat radiation transfer path. Each node loads electrical characteristics and real-time infrared thermal imaging temperature distribution data from the InfluxDB storage feature library. The weights are calculated based on the electrical correlation strength and heat exchange efficiency between adjacent nodes. The features of neighboring nodes are aggregated layer by layer and their own states are updated to generate a full-field fault probability distribution map, which can accurately locate the location of anomalies.

8. The method for analyzing photovoltaic power station equipment operation data based on the Internet of Things as described in claim 7, characterized in that, In S105, based on the output fault code and lifetime prediction value, the string dynamic reconstruction algorithm is initiated to replan the MPPT tracking path. For strings with detected PID risk, nighttime reverse bias repair is automatically triggered, including: Based on the fault code, locate the abnormal component ID, avoid components in the shadow / hot spot area, re-constrain the string electrical topology, and re-plan the string electrical topology. The constraints include the upper limit of single string voltage, the current imbalance between strings and the physical wiring distance limit. Relay switch matrix instructions are generated and reconstructed through the smart junction box. Differentiated repairs are implemented based on relay switch matrix instruction reconstruction. For low-to-medium risk, a fixed bias voltage of -200V / 2 hours is applied, while for high-risk, dynamic bias voltage adjustment and intelligent duration control are adopted. The optimal resistance value is dynamically calculated by combining the insulation resistance decay rate of the InfluxDB storage feature library. The leakage current change is monitored in real time during the repair process, and the operation is automatically executed at night. After the repair, the verification data is sent back to update the equipment health record.

9. A photovoltaic power station equipment operation data analysis system based on the Internet of Things, characterized in that, The data analysis system is used to perform the photovoltaic power plant equipment operation data analysis method according to any one of claims 1-8, the data analysis system comprising: The multi-source data acquisition module is used to collect current, voltage, temperature, irradiance, and ambient humidity data in real time through sensor nodes of photovoltaic modules / combiner boxes / inverters / weather stations, and supports hybrid networking transmission of LoRaWAN and ZigBee; The three-level verification module is used to perform outlier removal based on the Laida criterion, linear interpolation data compensation, wavelet denoising and temperature drift calibration, and output preprocessed data. The feature calculation engine module is used to extract electrical / thermal / environmental / temporal four-dimensional feature sets in parallel by device ID, build the InfluxDB time series database, and implement hierarchical management of hot and cold data. The hierarchical diagnostic module is used to identify basic faults using random forest, locate fault points by fusing infrared data with GCN, dynamically calculate the degradation index and output the lifetime prediction value. The closed-loop control module is used to perform string dynamic reconstruction, PID reverse bias repair, AGC / AVC instruction rolling optimization, and to store decision logs through blockchain.

10. The IoT-based photovoltaic power station equipment operation data analysis system as described in claim 9, characterized in that, The feature calculation engine module includes: Electrical characteristic unit, used to calculate the inflection point slope of IV curve and ripple spectrum entropy value; Thermal feature units are used to generate component temperature gradient field distribution maps. An environmental coupling unit is used to construct the irradiance-power response transfer function; The timing feature unit is used to extract the frequency domain abrupt change of the reverse current at night. The database construction unit is used to process and merge the acquired four-dimensional feature set to build the InfluxDB time series database and implement hierarchical management of hot and cold data.

Images