Photovoltaic power station operation state monitoring management system and method

Abstract

The invention discloses a photovoltaic power station operation state monitoring management system and method, and relates to the technical field of power systems and automatic control, and the method comprises the steps that a dual-mode temperature monitoring module is combined with a ceramic packaging contact sensor and a multispectral filtering non-contact sensor, and a correction temperature value is output through adaptive weight fusion; the environment adaptive calibration unit dynamically corrects the drifting of the sensor based on a CNN-BiLSTM model; the edge computing node detects early symptoms of the hot spots in real time and uploads abnormal data preferentially; the multi-protocol gateway can automatically switch communication channels, and is matched with LZW compression and CRC32 verification to guarantee data transmission; the above modules establish cooperative interaction through an industrial Ethernet, an RS485 bus and the like, and are in bidirectional communication with a cloud management platform. The temperature monitoring precision in a severe environment is improved, the hot spot response time is shortened, the communication bottleneck in a remote area is solved, and stable operation of a power station is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of power system and automation control technology, specifically to a photovoltaic power plant operation status monitoring and management system and method. Background Technology

[0002] Globally, photovoltaic (PV) installations are growing rapidly. As of 2023, my country's cumulative installed PV capacity exceeded 1000 GW, with distributed PV accounting for over 40%. Power plants must possess higher levels of intelligent management capabilities to meet grid connection standards and carbon emission accounting requirements.

[0003] Large-scale ground-mounted power plants cover areas of tens of square kilometers, and distributed photovoltaic equipment is scattered, making traditional manual inspections insufficient to meet real-time monitoring needs. For example, drone inspections can cover an area of ​​up to 50 hectares in a single operation, which is 20 times more efficient than manual inspections.

[0004] Operation and maintenance costs account for 15% to 20% of the total life cycle cost of a power plant, while equipment failures can lead to a 5% to 10% loss in annual power generation. Intelligent monitoring systems can reduce unplanned downtime and extend equipment life by more than 20% through predictive maintenance.

[0005] The volatility and randomness of distributed photovoltaic (PV) power generation impact grid power quality, causing issues such as harmonics and voltage fluctuations. The system needs to integrate anti-islanding protection and power quality monitoring functions to ensure grid connection safety.

[0006] Therefore, there is an urgent need for a photovoltaic power plant operation status monitoring and management system and method. In the face of sensor and communication bottlenecks, including insufficient environmental adaptability, traditional sensors are prone to accuracy drift under harsh conditions such as high temperature and high humidity. The component temperature monitoring response time is as long as 3 to 5 seconds, which makes it impossible to capture early signs of hot spots in time. Communication delays and interruptions occur because network coverage is weak in remote areas and 4G / 5G signals are easily interfered with by terrain, resulting in data transmission delays of more than 10 seconds, and even packet loss rates of up to 5%. Summary of the Invention

[0007] This invention aims to provide a photovoltaic power plant operation status monitoring and management system and method. It uses dual-mode temperature monitoring and dynamic calibration of CNN-BiLSTM model to detect hot spots in real time with string-level edge computing. By using multi-protocol redundant communication, it solves the problems of sensor accuracy drift, early warning lag, and communication delay and packet loss in remote areas under high temperature and humidity.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A photovoltaic power plant operation status monitoring and management system includes:

[0010] Dual-mode temperature monitoring module: integrates a contact thermopile sensor and a non-contact infrared sensor, deployed on the backsheet of photovoltaic modules and key electrical nodes; the contact thermopile sensor adopts ceramic encapsulation and aluminum nitride substrate; the non-contact infrared sensor suppresses environmental radiation interference through multispectral filtering technology; the dual-mode data outputs a corrected temperature value through an adaptive weighted fusion algorithm.

[0011] Edge computing nodes: equipped with multi-core heterogeneous processors and deployed at the photovoltaic string level; execute early hot spot detection algorithms in real time and trigger early warnings based on the component temperature gradient change rate; support local data caching and abnormal data priority upload mechanism;

[0012] Multi-protocol redundant communication gateway: integrates LoRa self-organizing network module, low-orbit satellite communication module and power line carrier communication module; the multi-protocol redundant communication gateway has a built-in signal strength monitoring unit, which automatically switches to satellite and PLC channels when the 4G / 5G signal strength is less than a predetermined threshold; it adopts LZW fragmented compression transmission protocol to split the preprocessed data into data packets of predetermined size, and ensures integrity through CRC32 check, and the retransmission timeout time is dynamically adjusted according to the link quality;

[0013] Environmental adaptive calibration unit: It has a built-in temperature and humidity compensation sensor array to collect environmental parameters in real time; the environmental adaptive calibration unit is equipped with a deep learning model based on CNN-BiLSTM, which takes the recent time series data of environmental parameters as input and outputs the drift prediction value of the dual-mode temperature monitoring module; the environmental adaptive calibration unit and the dual-mode temperature monitoring module are linked through the I²C bus to dynamically correct the detection data of the contact / non-contact sensors.

[0014] Inter-module interaction: The dual-mode temperature monitoring module is connected to the edge computing node via an industrial Ethernet; the edge computing node communicates with the multi-protocol redundant communication gateway via an RS485 bus; the multi-protocol redundant communication gateway establishes bidirectional communication with the cloud management platform, supporting the adjustment of communication protocol priority and data transmission frequency according to cloud instructions.

[0015] Preferably, the dual-mode temperature monitoring module specifically includes:

[0016] The contact thermopile sensor unit adopts a zirconia-toughened ceramic encapsulation shell and integrates a miniature thermopile array with aluminum nitride as the substrate; the non-contact infrared sensor unit has a built-in dual-band filter and suppresses interference from direct solar radiation and environmental thermal radiation through multispectral differential operation; the thermopile sensor and the infrared sensor are installed in a coaxial embedded structure at the monitoring point on the back sheet of the photovoltaic module.

[0017] Preferably, the dual-mode temperature monitoring module specifically includes:

[0018] The adaptive weighted fusion algorithm execution steps include: collecting relative humidity and ambient temperature around the monitoring point through a temperature and humidity compensation sensor array, dynamically calculating weight coefficients and outputting corrected temperature values.

[0019] Preferably, the edge computing node specifically includes:

[0020] The multi-core heterogeneous processor architecture uses an ARM Cortex-M7 core to perform data acquisition and cache management, and FPGA programmable logic units to execute the hot spot detection algorithm in parallel.

[0021] The hot spot early sign detection algorithm calculates the temperature gradient change rate of the component surface in real time. When the temperature gradient change rate is greater than or equal to a predetermined threshold for multiple consecutive sampling cycles, a red warning is triggered; when the temperature gradient change rate is within the predetermined threshold range, a yellow warning is triggered and high-frequency sampling is started.

[0022] The system employs a tiered caching and upload mechanism, with a local two-level cache: a circular buffer for storing regular data and a dedicated flash memory area for storing abnormal data. Once communication is restored, abnormal data is uploaded first, while data is saved in the order of red alert, yellow alert, and then regular data when transmission is interrupted.

[0023] Preferably, the multi-protocol redundancy communication gateway specifically includes:

[0024] Tri-mode communication unit: including LoRa self-organizing network module, low-Earth orbit satellite communication module, and power line carrier communication module;

[0025] The signal strength dynamic switching unit monitors the 4G / 5G signal strength in real time. When the strength is less than a predetermined threshold and reaches a duration threshold, channel switching is triggered. The switching priorities are satellite channel, PLC channel, and LoRa channel, respectively.

[0026] The link quality assessment matrix is ​​based on a scoring model constructed from packet loss rate, latency jitter, and bandwidth utilization, and a scoring formula is calculated.

[0027] Preferably, the multi-protocol redundancy communication gateway specifically includes:

[0028] The LZW fragmented compression transmission protocol execution steps include: a data preprocessing stage, in which the original data is compressed using the LZW algorithm, and the compressed data is split into fixed 256-byte data packets with an additional 4-byte CRC32 checksum;

[0029] During the dynamic retransmission control phase, the initial retransmission timeout is set to 200ms × the reciprocal of the LQE score; if a single packet fails to retransmit more than or equal to the predetermined number of times, the packet is marked as a non-critical data packet and the skip mechanism is activated.

[0030] During the satellite transmission optimization phase, the BeiDou short message pass-through mode is enabled for red alert data, and the alarm information is encoded into a 32-byte BeiDou-specific format, abandoning CRC check.

[0031] Preferably, the environment adaptive calibration unit specifically includes:

[0032] The environmental adaptive calibration unit includes: a distributed temperature and humidity compensation sensor array, consisting of SHT45 digital sensors arranged in a 4×4 matrix; real-time acquisition of temperature, relative humidity, and air pressure data at monitoring points;

[0033] The CNN-BiLSTM deep learning model receives 72 hours of continuous environmental parameter time-series data in its input layer; the output layer generates drift predictions from dual-mode sensors, which are divided into contact and non-contact types.

[0034] The I²C bus dynamic correction module sends drift compensation commands to the dual-mode temperature monitoring module according to the calibration cycle.

[0035] Preferably, the environment adaptive calibration unit specifically includes:

[0036] The CNN-BiLSTM deep learning model includes a CNN feature extraction layer and a BiLSTM temporal analysis layer; it calculates the drift prediction formula to obtain the drift value.

[0037] The calibration trigger conditions are as follows: real-time calibration is initiated when the predicted drift value exceeds a predetermined threshold or when the sudden change in ambient humidity exceeds a predetermined threshold; zero-point calibration is automatically performed at night, and sensor readings are compared with those of a standard blackbody source under no-irradiation conditions.

[0038] Preferably, the interaction relationships between the modules specifically include:

[0039] The cloud management platform's two-way control includes: dynamic adjustment of communication protocol priority; receiving protocol weight vectors from the cloud; forcibly enabling satellite channel transmission of early warning data when the weight value of the satellite communication protocol exceeds a predetermined threshold; adjusting the sampling frequency based on the network congestion index calculated by the cloud; and automatically switching to the fiber optic redundant channel when a predetermined number of consecutive error frames are detected on the RS485 bus.

[0040] Furthermore, a method for monitoring and managing the operational status of a photovoltaic power plant includes:

[0041] Raw temperature data is collected by a dual-mode temperature monitoring module deployed on the backsheet of photovoltaic modules and key electrical nodes; the dual-mode temperature monitoring module simultaneously activates a contact thermopile sensor and a non-contact infrared sensor.

[0042] Based on the environmental parameters collected in real time by the temperature and humidity compensation sensor array, the drift trend of the dual-mode temperature monitoring module is predicted by the CNN-BiLSTM deep learning model, and a drift correction coefficient is generated periodically. The drift correction coefficient is injected into the dual-mode raw temperature data, and the corrected temperature value is output through an adaptive weighted fusion algorithm.

[0043] After receiving the corrected temperature value, the string-level edge computing node calculates the temperature gradient change rate of the component surface in real time. When the change rate is greater than a predetermined threshold and lasts for 3 sampling cycles, it is determined to be an early sign of hot spot and triggers a local warning. At the same time, the 3σ criterion is applied to the collected data for anomaly filtering, and the temperature variance and current fluctuation coefficient are extracted as feature parameters. Regular data is compressed and cached locally, and abnormal data is marked as the highest priority.

[0044] The multi-protocol redundant communication gateway monitors the 4G / 5G signal strength in real time. When the signal strength is less than or equal to a predetermined threshold and lasts for a predetermined duration, it automatically switches to LoRa self-organizing network, low-orbit satellite, or power line carrier communication channel.

[0045] The transmitted data is fragmented and compressed using the LZW algorithm, and the fragmented data packets are appended with CRC32 check codes. During transmission, a mechanism is used to prioritize abnormal data and allocate a predetermined proportion of bandwidth, while regular data is transmitted periodically. When communication is interrupted, a 16GB local cache is used to temporarily store the data, and after recovery, the data is retransmitted based on the ARQ adaptive retransmission algorithm.

[0046] After receiving the transmitted data, the cloud management platform combines historical operating curves and environmental parameters to establish a hot spot development trend model and outputs component health scores and maintenance priority suggestions.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] In the temperature monitoring stage, the dual-mode temperature monitoring module combines the characteristics of contact and non-contact sensors. The contact sensor uses ceramic encapsulation and an aluminum nitride substrate to improve structural stability and heat conduction efficiency in high-temperature environments, avoiding accuracy drift caused by high temperatures in traditional sensors. The non-contact sensor relies on multispectral filtering technology to effectively suppress interference from solar radiation, environmental reflections, etc. Combined with the CNN-BiLSTM deep learning model of the environmental adaptive calibration unit, it can dynamically predict the sensor drift trend based on temperature and humidity time series data, and correct the detection data in real time through the I²C bus, significantly reducing measurement errors in high-temperature and high-humidity environments, ensuring accurate and reliable calibration temperature values, and providing a high-quality data foundation for hot spot detection.

[0049] In terms of data processing and early warning, edge computing nodes are deployed at the string level. By leveraging the real-time temperature gradient change rate of multi-core heterogeneous processors, they can quickly capture early signs of hot spots and trigger local early warnings, solving the 3-5 second response delay problem caused by traditional systems relying on cloud analysis. At the same time, through local caching and a mechanism that prioritizes the uploading of abnormal data, invalid data transmission is reduced, improving system response efficiency.

[0050] At the communication level, the multi-protocol redundant communication gateway integrates multiple channels of LoRa, satellite, and power line carrier, and can automatically switch communication links according to the 4G / 5G signal strength to adapt to complex terrain scenarios in remote areas. Combined with LZW fragmentation compression, CRC32 check and dynamic retransmission timeout adjustment, it effectively reduces data transmission latency and packet loss rate, solving the problem of latency exceeding 10 seconds and packet loss rate of up to 5% caused by signal interference in traditional communication.

[0051] In addition, the modules establish a stable connection through industrial Ethernet and RS485 bus. With the help of the cloud bidirectional communication mechanism, the protocol priority and data transmission frequency can be flexibly adjusted to adapt to the operation and maintenance needs of photovoltaic power plants of different scales. This improves the system's environmental adaptability, data reliability and operation and maintenance flexibility, providing a strong guarantee for the efficient and stable operation of the power plant. Attached Figure Description

[0052] Figure 1 This is an internal framework diagram of a photovoltaic power plant operation status monitoring and management system.

[0053] Figure 2 This is a flowchart of a photovoltaic power plant operation status monitoring and management method. Detailed Implementation

[0054] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0055] Reference Figure 1 As shown, a photovoltaic power plant operation status monitoring and management system includes:

[0056] Dual-mode temperature monitoring module: integrates a contact thermopile sensor and a non-contact infrared sensor, deployed on the backsheet of photovoltaic modules and key electrical nodes; the contact thermopile sensor adopts ceramic encapsulation and aluminum nitride substrate; the non-contact infrared sensor suppresses environmental radiation interference through multispectral filtering technology; the dual-mode data outputs a corrected temperature value through an adaptive weighted fusion algorithm.

[0057] Edge computing nodes: equipped with multi-core heterogeneous processors and deployed at the photovoltaic string level; execute early hot spot detection algorithms in real time and trigger early warnings based on the component temperature gradient change rate; support local data caching and abnormal data priority upload mechanism;

[0058] Multi-protocol redundant communication gateway: integrates LoRa self-organizing network module, low-orbit satellite communication module and power line carrier communication module; the multi-protocol redundant communication gateway has a built-in signal strength monitoring unit, which automatically switches to satellite and PLC channels when the 4G / 5G signal strength is less than a predetermined threshold; it adopts LZW fragmented compression transmission protocol to split the preprocessed data into data packets of predetermined size, and ensures integrity through CRC32 check, and the retransmission timeout time is dynamically adjusted according to the link quality;

[0059] Environmental adaptive calibration unit: It has a built-in temperature and humidity compensation sensor array to collect environmental parameters in real time; the environmental adaptive calibration unit is equipped with a deep learning model based on CNN-BiLSTM, which takes the recent time series data of environmental parameters as input and outputs the drift prediction value of the dual-mode temperature monitoring module; the environmental adaptive calibration unit and the dual-mode temperature monitoring module are linked through the I²C bus to dynamically correct the detection data of the contact / non-contact sensors.

[0060] Inter-module interaction: The dual-mode temperature monitoring module is connected to the edge computing node via an industrial Ethernet; the edge computing node communicates with the multi-protocol redundant communication gateway via an RS485 bus; the multi-protocol redundant communication gateway establishes bidirectional communication with the cloud management platform, supporting the adjustment of communication protocol priority and data transmission frequency according to cloud instructions.

[0061] It should be noted that the dynamic drift suppression chain works as follows: the environmental calibration unit captures microclimate differences through a 4×4 temperature and humidity matrix → the CNN-BiLSTM model outputs drift prediction values ​​→ the I²C bus provides millisecond-level instructions to correct the dual-mode sensor data → the fused data accuracy reaches ±0.3℃, solving the problem of inaccurate temperature measurement caused by high temperature and high humidity.

[0062] Zero-point self-calibration mechanism: Automatically compare sensor readings with standard blackbody source readings at night, and update model parameters by combining historical drift data, with annual cumulative drift controlled at <0.8℃;

[0063] Real-time edge protection layer: The string-level nodes calculate the temperature gradient change rate ΔT / Δt with a delay of ≤50ms. When the temperature gradient exceeds 1.5℃ / s for 3 consecutive cycles, high-frequency infrared thermal imaging is activated for re-judgment (10Hz). The temperature rise trend is predicted based on the heat capacity model. If necessary, the string is cut off within 30 seconds.

[0064] Cloud-based strategy optimization layer: Receives data from the entire network to train a component attenuation diagnosis model, pushes a lightweight version to edge nodes daily, and achieves real-time diagnosis of string current imbalance >15%. Communication protocol weight vector W = [W 4G W sat W PLC Dynamic adjustment of ];

[0065] In the formula, W represents the protocol weight vector, which is an array containing the weight values ​​of different communication protocols; W 4G W represents the weight value of the 4G communication protocol. sat This indicates the weight value of the satellite communication protocol. If the weight value of the satellite communication protocol is greater than 0.7, the satellite channel transmission of early warning data will be forcibly initiated; W PLC This indicates the weight value of the PLC (Power Line Communication) protocol;

[0066] Multimodal communication architecture:

[0067] Edge node, RS485 bus → communication gateway;

[0068] Communication gateway → 4G / 5G main channel → cloud;

[0069] Communication gateway → satellite backup channel → cloud;

[0070] Communication gateway → PLC backup channel communication gateway → Cloud.

[0071] The dual-mode temperature monitoring module specifically includes:

[0072] The contact thermopile sensor unit adopts a zirconia-toughened ceramic encapsulation shell and integrates a miniature thermopile array with aluminum nitride as the substrate; the non-contact infrared sensor unit has a built-in dual-band filter and suppresses interference from direct solar radiation and environmental thermal radiation through multispectral differential operation; the thermopile sensor and the infrared sensor are installed in a coaxial embedded structure at the monitoring point on the back sheet of the photovoltaic module.

[0073] The adaptive weighted fusion algorithm execution steps include: collecting relative humidity and ambient temperature around the monitoring point through a temperature and humidity compensation sensor array, dynamically calculating weight coefficients and outputting corrected temperature values.

[0074] It should be noted that the contact thermopile sensor:

[0075] Zirconia toughened ceramic encapsulation uses yttrium-stabilized zirconia (Y-TZP) material, with a fracture toughness of 8 MPa·m¹ / ² (compared to only 4 MPa·m¹ / ² for ordinary alumina ceramics). Its crack resistance is improved by 300% under 150℃ thermal cycling, solving the thermal expansion mismatch problem of traditional metal encapsulation.

[0076] The ceramic shell and the aluminum nitride substrate are filled with gold-tin eutectic solder (Au80Sn20), which has a thermal conductivity of ≥200W / (m·K), reducing the temperature measurement response time to 0.6 seconds;

[0077] The aluminum nitride substrate micro thermopile array integrates 256 pairs of Bi2Te3 / Sb2Te3 thermocouples on a 10×10mm substrate, with a Seebeck coefficient ≥450μV / K, and can detect micro temperature differences in the range of 0.05℃. The cold end of the thermopile has a built-in PT1000 thin film resistor, which is driven by a constant current source to compensate for ambient temperature drift in real time.

[0078] Non-contact infrared sensor:

[0079] Dual-band filtering: Band 1 (3-5μm) is for matching the peak radiation of the silicon-based backsheet of the photovoltaic module, with a transmittance of >90%; Band 2 (8-14μm) is for suppressing solar shortwave radiation (0.3-2.5μm) and environmental thermal noise (>14μm).

[0080] Difference operations, according to the formula Eliminate interference signals;

[0081] In the formula, It represents the voltage difference, that is, the difference between two voltage values; and These are two coefficients used for adjustment. and The weights; and These represent voltage values ​​for two different frequency bands; the specific values ​​of k1 and k2 were determined experimentally, and their ratio is 1.32 to 0.78.

[0082] Coaxial embedded structure: The angle between the thermopile and the optical axis of the infrared sensor is ≤0.5°, and the overlap of the detection area is >95%, which solves the temperature measurement deviation caused by misaligned installation; The sensor cavity is filled with aerogel thermal insulation material (thermal conductivity 0.02W / (m·K)), which reduces external heat conduction interference by 98%.

[0083] Adaptive weighted fusion algorithm:

[0084] The dynamic calculation model for weighting coefficients requires input parameters including relative humidity (RH) (0%–100%) and ambient temperature (T). env (-40~85℃), solar radiation intensity G (0-1500W / m²);

[0085] Weighting formula:

[0086]

[0087] In the formula, Infrared weight represents the proportion of infrared factors in the weight allocation; Heat weight represents the proportion of heat factors in the weight allocation; yes The complements of the two numbers, which sum to 1, under high temperature and high humidity conditions. Approaching 0.8, low-temperature drying conditions It tends towards 0.9;

[0088] Environmental parameter compensation mechanism: Humidity-induced heat capacity effect. When RH > 85%, the water film on the component surface forms additional heat capacity, resulting in lag in contact temperature measurement → the algorithm automatically increases the infrared weight.

[0089] The superposition interference of solar radiation is dynamically corrected for infrared thermometry values ​​using the G-value:

[0090]

[0091] In the formula, This is the original infrared thermometer value; This is the corrected infrared thermometer value, representing the temperature value dynamically corrected based on the solar radiation intensity G; it takes effect when G > 800 W / m².

[0092] The edge computing nodes specifically include:

[0093] The multi-core heterogeneous processor architecture uses an ARM Cortex-M7 core to perform data acquisition and cache management, and FPGA programmable logic units to execute the hot spot detection algorithm in parallel.

[0094] The hot spot early sign detection algorithm calculates the temperature gradient change rate of the component surface in real time. When the temperature gradient change rate is greater than or equal to a predetermined threshold for multiple consecutive sampling cycles, a red warning is triggered; when the temperature gradient change rate is within the predetermined threshold range, a yellow warning is triggered and high-frequency sampling is started.

[0095] The system employs a tiered caching and upload mechanism, with a local two-level cache: a circular buffer for storing regular data and a dedicated flash memory area for storing abnormal data. Once communication is restored, abnormal data is uploaded first, while data is saved in the order of red alert, yellow alert, and then regular data when transmission is interrupted.

[0096] It should be noted that multi-core heterogeneous processors:

[0097] The ARM Cortex-M7 core uses a dual DMA channel to directly connect to the ADC, transferring sampled data to memory with zero copying and a latency of ≤50μs; it also allocates a 2MB PSRAM high-speed cache to support 5Hz full temperature data temporary storage (covering the risk of 200ms communication interruption).

[0098] The low-power scheduling strategy switches to Stop2 mode (power consumption 0.8μA) without warning, and wakes up via RTC event with a wake-up cycle of ≤3ms; when communication is interrupted, the local decision-making power-saving mode is activated, consuming only 25% of the battery capacity within 72 hours.

[0099] FPGA parallel computing: Raw temperature data → FPGA preprocessing pipeline → gradient calculation engine → early warning decision tree;

[0100] Hardware acceleration for hot spot detection is achieved by embedding a 5-stage pipeline architecture within the FPGA: data normalization → 3×3 spatial convolution → temporal difference calculation → threshold comparison → early warning code generation.

[0101] Dynamic reconstruction allows for online updates of convolution kernel coefficients via a bitstream configuration file distributed from the cloud, supporting switching between operators such as Sobel and Laplace.

[0102] Algorithm for early signs of hot spots:

[0103] Table 1: Multi-level determination of temperature gradient change rate (ΔT / Δt):

[0104] Warning level Triggering conditions Response Action Physical meaning Red Alert ΔT / Δt ≥ 1.5℃ / s and continuously for ≥ 3 periods Start-up string power failure + BeiDou short message pass-through Hot spots have formed and are spreading rapidly. Yellow Alert 1.0℃ / s≤ΔT / Δt<1.5℃ / s Infrared thermal imaging re-judgment + data stored in a separate flash memory area Potential hot spot risk requires manual confirmation. Baseline monitoring ΔT / Δt<1.0℃ / s Data is stored in a circular buffer, and the sampling rate is reduced to 0.2Hz. Energy-saving operation under normal working conditions

[0105] Enhanced anti-false alarm capability and spatial correlation verification: When a single point triggers an alarm, it scans the data of 8 surrounding sensors. If ≥2 adjacent points ΔT / Δt>0.8℃ / s, the alarm is confirmed.

[0106] Environmental interference filtering is performed by introducing correction factors for solar radiation intensity G and wind speed V:

[0107]

[0108] In the formula, This is the corrected rate of temperature change; The original temperature change rate is denoted by G; solar radiation intensity is watts per square meter (W / m²); and wind speed is denoted by V. The effect is achieved when G > 1000 W / m² or V < 1 m / s, suppressing false positives caused by sudden cloud changes / calm winds.

[0109] Two-level storage area:

[0110] A circular buffer with a capacity of 8MB is used for circular storage, dedicated to storing routine monitoring data (sampling rate 0.2Hz); ECC-checked SDRAM is used to resist single-event upsets.

[0111] Independent flash memory area, using SLCNAND flash memory (100,000 erase / write cycles), stored in blocks according to warning level:

[0112] Red alert, raw temperature data + thermal image (20MB / event);

[0113] Yellow alert, gradient change curve + environmental parameters (2MB / event).

[0114] The multi-protocol redundant communication gateway specifically includes:

[0115] Tri-mode communication unit: including LoRa self-organizing network module, low-Earth orbit satellite communication module, and power line carrier communication module;

[0116] The signal strength dynamic switching unit monitors the 4G / 5G signal strength in real time. When the strength is less than a predetermined threshold and reaches a duration threshold, channel switching is triggered. The switching priorities are satellite channel, PLC channel, and LoRa channel, respectively.

[0117] The link quality assessment matrix is ​​based on a scoring model constructed from packet loss rate, latency jitter, and bandwidth utilization, and a scoring formula is calculated.

[0118] The LZW fragmented compression transmission protocol execution steps include: a data preprocessing stage, in which the original data is compressed using the LZW algorithm, and the compressed data is split into fixed 256-byte data packets with an additional 4-byte CRC32 checksum;

[0119] During the dynamic retransmission control phase, the initial retransmission timeout is set to 200ms × the reciprocal of the LQE score; if a single packet fails to retransmit more than or equal to the predetermined number of times, the packet is marked as a non-critical data packet and the skip mechanism is activated.

[0120] During the satellite transmission optimization phase, the BeiDou short message pass-through mode is enabled for red alert data, and the alarm information is encoded into a 32-byte BeiDou-specific format, abandoning CRC check.

[0121] It should be noted that the tri-mode communication unit is as follows:

[0122] The low-orbit satellite communication module (Orbcomm system) uses an omnidirectional slotted antenna (gain 5dBi) to support communication at elevation angles of 10° and solves the problem of mountainous terrain obstruction.

[0123] Power line carrier (PLC) module, operating frequency band 150-500kHz, coupling efficiency >92% (compatible with photovoltaic inverters).

[0124] LoRa self-organizing network module, SF=10 spreading factor, receiver sensitivity -148dBm, coverage radius of 3.2km in hilly environments; relay hop count ≤3 levels, end-to-end delay <800ms.

[0125] Table 2: Dynamic signal strength switching, dual threshold triggering mechanism:

[0126] Switching conditions Threshold setting Action response Main channel (4G / 5G) strength <-92dBm for ≥8 seconds Initiate satellite channel switching (priority 1) Satellite channel packet loss rate >12% for 5 consecutive times Degrade to PLC channel (priority 2) PLC carrier signal-to-noise ratio <15dB Enable LoRa self-organizing network (priority 3)

[0127] Seamless data bridging, with 128KB of data cached in local RAM before switching (overwriting 3-second interrupt).

[0128] Link state pre-handshake: After the switch command is issued, the new channel ACK response is confirmed through in-band signaling (delay < 200ms).

[0129] Link Quality Assessment Matrix (LQE): A multi-dimensional scoring model with the following scoring formula:

[0130] In the formula, LQE is the link quality assessment score, which is a comprehensive score used to measure the overall quality of the link; Packet loss rate represents the proportion of data packets lost during data transmission, ranging from 0 to 1, where 0 indicates no packet loss and 1 indicates that all data packets are lost. In the scoring model, packet loss rate has the highest weight. Jitter is the fluctuation in the arrival time of data packets, measured in milliseconds. In PLCs and channels, jitter is an important indicator. Bandwidth utilization represents the ratio of currently available bandwidth to maximum bandwidth, ranging from 0 to 1. Maximum bandwidth indicates the maximum bandwidth that the link can reach;

[0131] Dynamic weight adjustment; when the ambient temperature is >70℃, increase Weighting to 0.7 (high temperature leads to increased bit error rate);

[0132] During heavy rain in mountainous areas, The weight was increased to 0.4 (rain attenuation exacerbated signal fluctuations).

[0133] LZW fragmented compression transmission protocol:

[0134] In the data preprocessing stage, a customized compression algorithm was adopted, using a 12-bit encoding dictionary to balance compression ratio and memory usage; δ encoding (adjacent value difference compression) was enabled for temperature data, increasing the compression ratio to 4:1;

[0135] Enhanced fragmentation strategy: 256-byte fixed fragmentation to match satellite MTU; each packet is appended with 4 bytes of CRC32, capable of detecting ≥32-bit errors;

[0136] Dynamic retransmission control mechanism with intelligent timeout calculation: RTO = 200ms × (1 / LQE)

[0137] When LQE=0.2, RTO=1 second (weak network tolerates long wait times).

[0138] When LQE=0.8, RTO=250ms (fast retransmission on high-quality links);

[0139] A skip mechanism is implemented for fault tolerance; packets that fail to retransmit more than twice are marked as non-critical packets; data is reconstructed at the receiving end using linear interpolation.

[0140] Satellite transmission is subject to extreme conditions, and BeiDou short message transmission rules apply, limiting it to red alert data (ΔT / Δt≥1.5℃ / s×3 cycles).

[0141] 32-byte encoding format:

[0142] [2-byte power plant ID][2-byte serial number][4-byte timestamp][3-byte longitude][3-byte latitude][2-byte component location][16-byte temperature gradient array];

[0143] Abandoning CRC, the BeiDou protocol itself contains BCH error correction code (12-bit error correction capability), with an additional CRC redundancy of >80%.

[0144] The environmental adaptive calibration unit specifically includes:

[0145] The environmental adaptive calibration unit includes: a distributed temperature and humidity compensation sensor array, consisting of SHT45 digital sensors arranged in a 4×4 matrix; real-time acquisition of temperature, relative humidity, and air pressure data at monitoring points;

[0146] The CNN-BiLSTM deep learning model receives 72 hours of continuous environmental parameter time-series data in its input layer; the output layer generates drift predictions from dual-mode sensors, which are divided into contact and non-contact types.

[0147] The I²C bus dynamic correction module sends drift compensation commands to the dual-mode temperature monitoring module according to the calibration cycle.

[0148] The CNN-BiLSTM deep learning model includes a CNN feature extraction layer and a BiLSTM temporal analysis layer; it calculates the drift prediction formula to obtain the drift value.

[0149] The calibration trigger conditions are as follows: real-time calibration is initiated when the predicted drift value exceeds a predetermined threshold or when the sudden change in ambient humidity exceeds a predetermined threshold; zero-point calibration is automatically performed at night, and sensor readings are compared with those of a standard blackbody source under no-irradiation conditions.

[0150] It should be noted that the distributed temperature and humidity compensation sensor array:

[0151] The 4×4 matrix layout captures microclimate differences. Sixteen SHT45 sensors are deployed in a grid pattern with a spacing of 25cm in a 1m² monitoring area. This can identify uneven heat dissipation in the back panel of the component (temperature difference > 0.8℃) and edge condensation effect (humidity gradient > 10%RH / m).

[0152] The design incorporates redundancy to resist occlusion. In the event of failure at any single point, the data can be reconstructed using an inverse distance-weighted interpolation algorithm with an error of <0.3℃.

[0153] The SHT45 sensor is adapted for extreme conditions and uses a polyimide protective film (dustproof rating IP6X). Its response time is ≤4 seconds in a high humidity environment of 85%RH. The barometric pressure measurement range is 300-1200hPa, which is used to correct for boiling point drift caused by altitude (correction factor +0.7% for every 1000 meters increase).

[0154] CNN-BiLSTM deep learning model:

[0155] Dual-channel neural network architecture: input data → CNN feature extraction;

[0156] Input data → BiLSTM timing analysis;

[0157] CNN feature extraction → feature fusion layer → drift prediction output;

[0158] The CNN feature extraction layer has 3 convolutional kernels (16 / 32 / 64 channels) to capture temperature and humidity abrupt spikes and day-night cycle patterns; the max pooling layer compresses the feature dimension to 1 / 8, reducing the computational load.

[0159] BiLSTM time-series analysis layer, bidirectional LSTM unit (128 hidden layers), learns 72 hours of environmental memory; outputs a time-delay effect model of environmental parameters on sensor drift;

[0160] Drift prediction formula, contact sensor drift model:

[0161] In the formula, This refers to the drift amount of the contact sensor. and These are the model parameters; ΔRH is the relative humidity change rate (% / h), reflecting the water vapor infiltration rate; The activation energy (0.35 eV) characterizes the thermal aging properties of the ceramic encapsulation material; k is the Boltzmann constant; and T is the absolute temperature (Kelvin).

[0162] Non-contact sensor drift model:

[0163] In the formula, This refers to the drift amount of a non-contact sensor. and Here are the model parameters; G is the solar radiation intensity (W / m²); θ is the lens contamination accumulation angle; The rate of change of ambient temperature (°C / h) causes thermal deformation of the lens.

[0164] Table 3: Dynamic calibration mechanism, intelligent triggering logic for real-time calibration:

[0165] Triggering conditions Threshold setting Calibration action Predicted drift value δ≥0.5℃ for 30 minutes Immediately send I²C compensation command Humidity sudden change ΔRH≥15% / h Initiate high-frequency sampling + model re-evaluation Nighttime zero-point calibration Irradiation intensity <10W / m² Comparison with blackbody source benchmark

[0166] I²C bus instruction set, instruction encoding format:

[0167] [1-byte header][1-byte sensor ID][2-byte temperature compensation value][2-byte humidity compensation value][1-byte CRC];

[0168] Anti-interference transmission, using 400kHz high-speed mode, single command transmission time ≤0.2ms; each frame is supplemented with BCH error correction code (can correct 2-bit errors).

[0169] Standard blackbody source zero-point calibration; blackbody source technical specifications: emissivity 0.99±0.005 (covering the 8-14μm band); temperature stability ±0.03℃ / h (constant current temperature control technology);

[0170] The calibration process involves a stable ambient temperature phase (≥30 minutes with no ±0.1℃ fluctuation); the distance between the sensor and the blackbody source is fixed at 50±0.5cm; 10 readings are collected and the median value is taken; if the deviation is >0.5℃, the model parameters are updated.

[0171] The specific interaction relationships between modules include:

[0172] The cloud management platform's two-way control includes: dynamic adjustment of communication protocol priority; receiving protocol weight vectors from the cloud; forcibly enabling satellite channel transmission of early warning data when the weight value of the satellite communication protocol exceeds a predetermined threshold; adjusting the sampling frequency based on the network congestion index calculated by the cloud; and automatically switching to the fiber optic redundant channel when a predetermined number of consecutive error frames are detected on the RS485 bus.

[0173] It should be noted that the communication protocol priority is dynamically adjusted:

[0174] The cloud-based generation logic of the protocol weight vector (W), which consists of three parts: , and These represent the weights of 4G communication, satellite communication, and power line communication, respectively:

[0175]

[0176] C represents the real-time communication cost (yuan / segment), and the satellite channel weight is negatively correlated with the cost.

[0177] L represents the regional network load (0-1), and satellite weight is increased when 4G is congested.

[0178] E represents the environmental risk level (levels 1-3), indicating the intensity of sandstorms / rainstorms. ≥0.8;

[0179] Threshold triggering mechanism, when When the red alert data is above 0.7, the data is directly transmitted to the satellite channel (skipping local decision-making), and the BeiDou short message dual-channel backup (satellite + LoRa transparent transmission) is activated.

[0180] Specialized strategies for mountain power stations include a terrain attenuation compensation algorithm and a pre-loaded communication blind zone map based on a digital elevation model (DEM). When the string coordinates fall into the blind zone... =0.9, forcing satellite as the primary channel. =0.1, with PLC as a supplement;

[0181] Electricity prices will automatically increase during off-peak hours (11:00 PM - 7:00 AM). Reduce costs to 0.8 by utilizing idle power line bandwidth.

[0182] Network congestion index-driven, quantification model of the congestion index (CI):

[0183]

[0184] In the formula, CI stands for Congestion Index, a comprehensive score used to measure the degree of network congestion; α, β, and γ are the weighting coefficients for packet loss rate, latency jitter, and bandwidth utilization, respectively, used to adjust their influence in the congestion index; Packet Loss is the packet loss rate, representing the proportion of data packets lost during data transmission, usually expressed as a percentage; Delay Jitter is the latency jitter, representing the fluctuation in data packet arrival time, in milliseconds (ms); and Bandwidth Utilization is the bandwidth utilization, representing the current network bandwidth usage, usually expressed as a percentage, ranging from 0 to 1.

[0185] Table 4: Dynamic Coefficient Setting (Based on Channel Type):

[0186] Channel type α β γ 4G / 5G 0.5 0.3 0.2 satellite 0.7 0.2 0.1 PLC 0.3 0.5 0.2

[0187] Table 5: Sampling frequency hierarchical control:

[0188] CI range Regulation strategy Application scenarios CI≤0.3 Full-frequency sampling (5Hz) Mandatory activation during red alert period 0.3<CI≤0.6 Reduce frequency to 2Hz, infrared thermal imaging off. Yellow alert + Moderate communication congestion CI>0.6 Only maintain a base temperature sampling rate of 0.2Hz. Backup monitoring during severe communication outages

[0189] RS485 bus fault, Level 3 fault diagnosis:

[0190] ≥3 consecutive CRC errors → trigger bus self-test;

[0191] If ≥2 frame synchronization failures occur during self-test, a physical layer fault is determined.

[0192] Immediately enable the 1310nm fiber optic redundant channel (switching delay <200ms).

[0193] Fiber optic channel characteristics: single-mode fiber transmission distance ≥20km (no repeater required); Manchester coding is used to resist electromagnetic interference.

[0194] The black box records and stores bus messages from 5 minutes prior to the fault. Key fields include: [timestamp], [voltage fluctuation value], [error mode], and [ambient temperature].

[0195] Cloud-based repair guidance matches similar cases based on the historical fault database and pushes repair solutions, such as: "2025-08-12 09:23: Impedance mutation detected (>120Ω), it is recommended to check the oxidation of the wiring terminals."

[0196] Reference Figure 2 As shown, a method for monitoring and managing the operating status of a photovoltaic power plant includes:

[0197] Raw temperature data is collected by a dual-mode temperature monitoring module deployed on the backsheet of photovoltaic modules and key electrical nodes; the dual-mode temperature monitoring module simultaneously activates a contact thermopile sensor and a non-contact infrared sensor.

[0198] Based on the environmental parameters collected in real time by the temperature and humidity compensation sensor array, the drift trend of the dual-mode temperature monitoring module is predicted by the CNN-BiLSTM deep learning model, and a drift correction coefficient is generated periodically. The drift correction coefficient is injected into the dual-mode raw temperature data, and the corrected temperature value is output through an adaptive weighted fusion algorithm.

[0199] After receiving the corrected temperature value, the string-level edge computing node calculates the temperature gradient change rate of the component surface in real time. When the change rate is greater than a predetermined threshold and lasts for 3 sampling cycles, it is determined to be an early sign of hot spot and triggers a local warning. At the same time, the 3σ criterion is applied to the collected data for anomaly filtering, and the temperature variance and current fluctuation coefficient are extracted as feature parameters. Regular data is compressed and cached locally, and abnormal data is marked as the highest priority.

[0200] The multi-protocol redundant communication gateway monitors the 4G / 5G signal strength in real time. When the signal strength is less than or equal to a predetermined threshold and lasts for a predetermined duration, it automatically switches to LoRa self-organizing network, low-orbit satellite, or power line carrier communication channel.

[0201] The transmitted data is fragmented and compressed using the LZW algorithm, and the fragmented data packets are appended with CRC32 check codes. During transmission, a mechanism is used to prioritize abnormal data and allocate a predetermined proportion of bandwidth, while regular data is transmitted periodically. When communication is interrupted, a 16GB local cache is used to temporarily store the data, and after recovery, the data is retransmitted based on the ARQ adaptive retransmission algorithm.

[0202] After receiving the transmitted data, the cloud management platform combines historical operating curves and environmental parameters to establish a hot spot development trend model and outputs component health scores and maintenance priority suggestions.

[0203] It should be noted that the dual-mode temperature sensing is as follows:

[0204] The physical redundancy design features a contact thermopile that is directly attached to the backplane of the module to capture microscopic hot spots; a non-contact infrared sensor (120° field of view) scans the macroscopic temperature field, and dual-source data avoids the risk of single-point failure.

[0205] Environmental interference decoupling: The temperature and humidity compensation array generates component-level microenvironment parameters through spatial interpolation algorithms, eliminating local temperature drift caused by mountain shadows and vegetation reflections;

[0206] CNN-BiLSTM drift prediction:

[0207] Multiphysics coupling modeling integrates 72 hours of environmental time-series data, sensor aging history, and component operating current curves as inputs, and outputs drift values ​​with a confidence interval of ±0.15℃.

[0208] Adaptive weighted fusion formula:

[0209] In the formula, The corrected temperature value is a weighted average of the measurements from contact and non-contact sensors; w is a weighting coefficient used to adjust the relative importance of the measurements from contact and non-contact sensors. The temperature value measured by the contact sensor; Temperature values ​​measured by non-contact sensors;

[0210] The weighting coefficient w is dynamically adjusted according to environmental conditions, and the calculation formula is as follows:

[0211] In the formula, ΔRH is the relative humidity change rate (% / h), reflecting the water vapor infiltration rate; G is the solar radiation intensity (W / m²), which affects the measurement accuracy of the non-contact sensor; The aging time of the sensor may affect the measurement accuracy of the sensor;

[0212] In high temperature and high humidity scenarios (>85%RH), the weight of the contact sensor is automatically reduced to 0.3 to suppress water vapor thermal conduction error.

[0213] Triple protection for edge computing nodes, hardware acceleration for hot spot recognition, FPGA-embedded 5-stage pipeline detection algorithm (spatial convolution → temporal difference → threshold comparison), completing 256-point temperature field scan in 0.3 seconds;

[0214] Intelligent upgrade for anomaly filtering, 3σ criterion for linkage current fluctuation coefficient When σ>2℃ and A value greater than 0.15 is considered a genuine anomaly, thus avoiding interference from cloud cover.

[0215] Table 6: Caching Tiered Strategy

[0216] Data types Storage strategy Capacity allocation Red Alert Independent SLC flash memory (100,000 erase / write cycles) 8GB Yellow Alert Circular buffer (circular overlay) 6GB Regular data Compressed storage (compression ratio 4:1) 2GB

[0217] Example:

[0218] Step 1: Dual-mode sensor installation. Drill a hole in the center of the photovoltaic module backsheet to install the ceramic-encapsulated thermopile sensor, fill it with aluminum nitride thermal paste, and ensure tight adhesion to the silicon substrate. Fix the infrared sensor 50cm above the module with a bracket, adjust the pitch angle to 32° (covering a 6×4 module matrix), and enable multispectral filtering to remove snow reflection interference. After powering on, perform dual-mode zero-point calibration: Under a no-irradiation environment at midnight, calibrate the dual-sensor deviation value with a blackbody radiation source (50℃±0.1℃) and record the initial compensation coefficient.

[0219] Step 2: Deploy the environmental array. Deploy one 16-channel temperature and humidity array (SHT45) for every 4 strings, with a sensor spacing of 2m×2m and a distance of 10cm from the backplane; connect the dual-mode sensors via I²C bus (shielded twisted pair, impedance matched to 120Ω), and configure the sampling rate to 1Hz;

[0220] Step 3: Continuously collect 72 hours of environmental time-series data (temperature / humidity / air pressure / irradiance), and simultaneously inject salt spray aging simulation signals (humidity >95% for 12 hours); train the CNN-BiLSTM deep learning model to output drift compensation values; among them, the contact sensor has positive drift and the infrared sensor has negative drift;

[0221] Step 4: Configure dynamic fusion rules and set basic weights in a normal temperature environment:

[0222] The response rules include when humidity > 85%, The weighting is reduced to 0.3 to suppress water vapor interference; when the irradiation is >1000W / m², The weighting was increased to 0.8 to avoid solar reflection errors; among them, The fused temperature value is a weighted average of measurements from both contact and non-contact sensors. The temperature value measured by the contact sensor; The temperature value is measured by a non-contact sensor;

[0223] Step 5: Edge node algorithm activation, hotspot detection pipeline programming, and solidification of the three-level hardware acceleration algorithm on a multi-core heterogeneous processor:

[0224] Spatial convolutional layer, 3×3 template scan temperature field, to identify local hotspots (resolution 0.1℃).

[0225] Time difference generator, calculates the temperature gradient ΔT / Δt over 5 seconds (accuracy 0.01℃ / s);

[0226] Threshold trigger: ΔT / Δt ≥ 1.5℃ / s for 3 consecutive cycles → triggers a red alert;

[0227] Configure 3σ-current-linked filtering; mark true anomalies when temperature variance > 2.5σ and current fluctuation > 18%.

[0228] Step 6: Enable tiered caching mechanism and allocate 16GB of flash memory:

[0229] Red alert for 8GB SLC block storage (100,000 erase / write cycles);

[0230] A yellow warning (circular overwrite) is present in the 6GBMLC circular buffer.

[0231] 2GB TLC storage for LZO compressed regular data (compression ratio ≥4:1);

[0232] Step 7: Multi-protocol channel integration testing:

[0233] Satellite module, register with Orbcomm network, set elevation angle ≥15° (to avoid obstruction by Kunlun Mountains);

[0234] PLC carrier, coupled to the inverter output with a 150kHz signal (impedance matched to 50Ω±5%).

[0235] LoRa network, configured with SF=10, channel 11 (868MHz), relay hop count ≤3;

[0236] Step 8: Set up the anti-damage transmission protocol. The fragmentation rules include 64 bytes of early warning data packets (including the BeiDou short message format header) and 256 bytes of regular data packets (incrementally encoded and stored ΔT).

[0237] Dynamic retransmission mechanism: RTO = 200ms × (1 / LQE), when LQE < 0.3, the maximum RTO is 1 second; maximum number of retransmissions = min(5, 3 × LQE);

[0238] Step 9: Industrial Ethernet extreme verification, 128 nodes concurrently transmitting Modbus TCP frames: [Temperature value: 4 bytes][Status code: 2 bytes][Timestamp: 4 bytes]; S485 / fiber redundancy switching, main channel baud rate 115200bps, daisy chain topology (≤32 nodes); simulated fault, injecting continuous CRC error frames, actual measured switching time to fiber channel 183ms (<200ms target);

[0239] Step 10: Dynamic communication scheduling is enabled, and protocol weight vectors are distributed from the cloud.

[0240] When a sandstorm warning is issued, W=[0.1,0.8,0.1] (satellite channel forced dominance);

[0241] During off-peak electricity pricing periods, W=[0.1,0.1,0.8] (prioritizing cost reduction for PLC channels);

[0242] Step 11: Training the health rating model, inputting historical data including peak records of temperature gradient change rate, decay slope of component IV curve, and spatiotemporal distribution map of hot spot events;

[0243] Finally, output the operation and maintenance instructions: calculate the health status using the health status formula (health status = 90 - 0.3 × (hot spot index) - 0.2 × (aging rate)); if the health status is <60 → re-inspect the drone within 48 hours; if the health status is <40 → emergency manual handling within 24 hours.

[0244] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A photovoltaic power plant operation status monitoring and management system, characterized in that, include: Dual-mode temperature monitoring module: integrates a contact thermopile sensor and a non-contact infrared sensor, deployed on the backsheet of photovoltaic modules and key electrical nodes; the contact thermopile sensor adopts ceramic encapsulation and aluminum nitride substrate; the non-contact infrared sensor suppresses environmental radiation interference through multispectral filtering technology; the dual-mode data outputs a corrected temperature value through an adaptive weighted fusion algorithm. Edge computing nodes: equipped with multi-core heterogeneous processors and deployed at the photovoltaic string level; execute early hot spot detection algorithms in real time and trigger early warnings based on the component temperature gradient change rate; support local data caching and abnormal data priority upload mechanism; Multi-protocol redundant communication gateway: integrates LoRa self-organizing network module, low-orbit satellite communication module and power line carrier communication module; the multi-protocol redundant communication gateway has a built-in signal strength monitoring unit, which automatically switches to satellite and PLC channels when the 4G / 5G signal strength is less than a predetermined threshold; it adopts LZW fragmented compression transmission protocol to split the preprocessed data into data packets of predetermined size, and ensures integrity through CRC32 check, and the retransmission timeout time is dynamically adjusted according to the link quality; Environmental adaptive calibration unit: It has a built-in temperature and humidity compensation sensor array to collect environmental parameters in real time; the environmental adaptive calibration unit is equipped with a deep learning model based on CNN-BiLSTM, which takes the recent time series data of environmental parameters as input and outputs the drift prediction value of the dual-mode temperature monitoring module; the environmental adaptive calibration unit and the dual-mode temperature monitoring module are linked through the I²C bus to dynamically correct the detection data of the contact / non-contact sensors. Inter-module interaction: The dual-mode temperature monitoring module is connected to the edge computing node via an industrial Ethernet; the edge computing node communicates with the multi-protocol redundant communication gateway via an RS485 bus; the multi-protocol redundant communication gateway establishes bidirectional communication with the cloud management platform, supporting the adjustment of communication protocol priority and data transmission frequency according to cloud instructions.

2. The photovoltaic power plant operation status monitoring and management system according to claim 1, characterized in that, The dual-mode temperature monitoring module specifically includes: The contact thermopile sensor unit adopts a zirconia-toughened ceramic encapsulation shell and integrates a miniature thermopile array with aluminum nitride as the substrate; the non-contact infrared sensor unit has a built-in dual-band filter and suppresses interference from direct solar radiation and environmental thermal radiation through multispectral differential operation; the thermopile sensor and the infrared sensor are installed in a coaxial embedded structure at the monitoring point on the back sheet of the photovoltaic module.

3. The photovoltaic power plant operation status monitoring and management system according to claim 1, characterized in that, The dual-mode temperature monitoring module specifically includes: The adaptive weighted fusion algorithm execution steps include: collecting relative humidity and ambient temperature around the monitoring point through a temperature and humidity compensation sensor array, dynamically calculating weight coefficients and outputting corrected temperature values.

4. The photovoltaic power plant operation status monitoring and management system according to claim 1, characterized in that, The environmental adaptive calibration unit specifically includes: The environmental adaptive calibration unit includes: a distributed temperature and humidity compensation sensor array, consisting of SHT45 digital sensors arranged in a 4×4 matrix; real-time acquisition of temperature, relative humidity, and air pressure data at monitoring points; The CNN-BiLSTM deep learning model receives 72 hours of continuous environmental parameter time-series data in its input layer; the output layer generates drift predictions from dual-mode sensors, which are divided into contact and non-contact types. The I²C bus dynamic correction module sends drift compensation commands to the dual-mode temperature monitoring module according to the calibration cycle.

5. The photovoltaic power plant operation status monitoring and management system according to claim 1, characterized in that, The environmental adaptive calibration unit specifically includes: The CNN-BiLSTM deep learning model includes a CNN feature extraction layer and a BiLSTM temporal analysis layer; it calculates the drift prediction formula to obtain the drift value. The calibration trigger conditions are as follows: real-time calibration is initiated when the predicted drift value exceeds a predetermined threshold or when the sudden change in ambient humidity exceeds a predetermined threshold; zero-point calibration is automatically performed at night, and sensor readings are compared with those of a standard blackbody source under no-irradiation conditions.

6. The photovoltaic power plant operation status monitoring and management system according to claim 1, characterized in that, The edge computing nodes specifically include: The multi-core heterogeneous processor architecture uses an ARM Cortex-M7 core to perform data acquisition and cache management, and FPGA programmable logic units to execute the hot spot detection algorithm in parallel. The hot spot early sign detection algorithm calculates the temperature gradient change rate on the component surface in real time. When the temperature gradient change rate is greater than or equal to a predetermined threshold for multiple consecutive sampling cycles, a red warning is triggered; when the temperature gradient change rate is within the predetermined threshold range, a yellow warning is triggered and high-frequency sampling is initiated. The system employs a tiered caching and upload mechanism, with a local two-level cache: a circular buffer for storing regular data and a dedicated flash memory area for storing abnormal data. Once communication is restored, abnormal data is uploaded first, while data is saved in the order of red alert, yellow alert, and then regular data when transmission is interrupted.

7. The photovoltaic power plant operation status monitoring and management system according to claim 1, characterized in that, The multi-protocol redundant communication gateway specifically includes: Tri-mode communication unit: including LoRa self-organizing network module, low-orbit satellite communication module, and power line carrier communication module; The signal strength dynamic switching unit monitors the 4G / 5G signal strength in real time. When the strength is less than a predetermined threshold and reaches a duration threshold, channel switching is triggered. The switching priorities are satellite channel, PLC channel, and LoRa channel, respectively. The link quality assessment matrix is ​​based on a scoring model constructed from packet loss rate, latency jitter, and bandwidth utilization, and a scoring formula is calculated.

8. The photovoltaic power plant operation status monitoring and management system according to claim 1, characterized in that, The multi-protocol redundant communication gateway specifically includes: The LZW fragmented compression transmission protocol execution steps include: a data preprocessing stage, in which the original data is compressed using the LZW algorithm, and the compressed data is split into fixed 256-byte data packets with an additional 4-byte CRC32 checksum; During the dynamic retransmission control phase, the initial retransmission timeout is set to 200ms × the reciprocal of the LQE score; if a single packet fails to retransmit more than or equal to the predetermined number of times, the packet is marked as a non-critical data packet and the skip mechanism is activated. During the satellite transmission optimization phase, the BeiDou short message pass-through mode is enabled for red alert data, and the alarm information is encoded into a 32-byte BeiDou-specific format, abandoning CRC check.

9. The photovoltaic power plant operation status monitoring and management system according to claim 1, characterized in that, The specific interaction relationships between modules include: The cloud management platform's two-way control includes: dynamic adjustment of communication protocol priority; receiving protocol weight vectors from the cloud; forcibly enabling satellite channel transmission of early warning data when the weight value of the satellite communication protocol exceeds a predetermined threshold; adjusting the sampling frequency based on the network congestion index calculated by the cloud; and automatically switching to the fiber optic redundant channel when a predetermined number of consecutive error frames are detected on the RS485 bus.

10. A method for monitoring and managing the operating status of a photovoltaic power station, characterized in that, include: Raw temperature data is collected by a dual-mode temperature monitoring module deployed on the backsheet of photovoltaic modules and key electrical nodes; The dual-mode temperature monitoring module simultaneously activates the contact thermopile sensor and the non-contact infrared sensor. Based on the environmental parameters collected in real time by the temperature and humidity compensation sensor array, the drift trend of the dual-mode temperature monitoring module is predicted by the CNN-BiLSTM deep learning model, and a drift correction coefficient is generated periodically. The drift correction coefficient is injected into the dual-mode raw temperature data, and the corrected temperature value is output through an adaptive weighted fusion algorithm. After receiving the corrected temperature value, the string-level edge computing node calculates the temperature gradient change rate of the component surface in real time. When the change rate is greater than a predetermined threshold and lasts for 3 sampling cycles, it is determined to be an early sign of hot spot and triggers a local warning. At the same time, the 3σ criterion is applied to the collected data for anomaly filtering, and the temperature variance and current fluctuation coefficient are extracted as feature parameters. Regular data is compressed and cached locally, and abnormal data is marked as the highest priority. The multi-protocol redundant communication gateway monitors the 4G / 5G signal strength in real time. When the signal strength is less than or equal to a predetermined threshold and lasts for a predetermined duration, it automatically switches to LoRa self-organizing network, low-orbit satellite, or power line carrier communication channel. The transmitted data is fragmented and compressed using the LZW algorithm, and the fragmented data packets are appended with CRC32 check codes. During transmission, a mechanism is used to prioritize abnormal data and allocate a predetermined proportion of bandwidth, while regular data is transmitted periodically. When communication is interrupted, a 16GB local cache is used to temporarily store the data, and after recovery, the data is retransmitted based on the ARQ adaptive retransmission algorithm. After receiving the transmitted data, the cloud management platform combines historical operating curves and environmental parameters to establish a hot spot development trend model and outputs component health scores and maintenance priority suggestions.

Images